Methods, devices and applications for retrieving genetic information

By combining large language models and scene-based intelligent agents, customized retrieval strategies are automatically generated, solving the problem of complex and highly specialized manual operations by users in existing technologies, and achieving efficient and accurate gene information retrieval.

CN123120064APending Publication Date: 2026-11-17SHENZHEN HUADA YONGSHENG INTELLIGENT TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510591295.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-08
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing gene sequence retrieval methods rely on manual operation by users, requiring users to have high levels of professional knowledge and technical skills. They are inefficient, complex, and prone to errors, especially under complex and ever-changing research needs.

Method used

It employs a large language model for intelligent interaction, generates customized search strategies through scene-based intelligent agents, automates gene information retrieval, reduces the technical threshold for users, and improves search efficiency.

Benefits of technology

It enables efficient gene information retrieval without requiring specialized knowledge, simplifies the operation process, improves retrieval efficiency and accuracy, and adapts to diverse research needs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120064A_ABST
    Figure CN123120064A_ABST
Patent Text Reader

Abstract

This application proposes a method, apparatus, and application for retrieving gene information. The method includes: receiving search prompts via a large language model, the search prompts containing prompts related to the target gene; determining the search target of the search prompts via the large language model, where different search targets have mapping relationships with different scene agents, and each scene agent encapsulates processing logic corresponding to the search target, the processing logic including a main database to be searched; generating a search strategy through the target scene agent with the mapping relationship to the search target, the search strategy including search keywords and a target main database, the target main database being the main database in the logic encapsulated by the target scene agent; and searching the target main database based on the search keywords to determine the search results. This method automatically generates customized search strategies for different search targets, effectively improving search efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to methods, apparatus and applications for retrieving gene information, and more specifically to methods and apparatus for retrieving gene information, electronic devices, computer-readable storage media and computer program products. Background Technology

[0002] Existing gene sequence retrieval methods rely on users accessing and filtering public databases, such as UniProt and NCBI, through web-based interfaces or command-line scripts. Users need to manually select relevant databases, enter search keywords, and complete the data retrieval through multiple clicks or script executions.

[0003] Traditional bioinformatics data retrieval methods rely on user-driven searches and filtering from public databases (such as NCBI and UniProt). This necessitates efficient data extraction and analysis using specific web search interfaces, command-line tools, or graphical user interface software. Users must possess strong knowledge of biological database information retrieval and bioinformatics. These methods are effective for simple, small-scale queries. However, they become cumbersome and inefficient when users' research needs are complex and varied, especially when multiple rounds of data filtering, complex data comparisons, and customized searches are required. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art. To this end, one objective of this application is to propose a method capable of automating the retrieval of gene information.

[0005] Specifically, the technical solution of this application is as follows:

[0006] In a first aspect of this application, a method for retrieving gene information is proposed. According to an embodiment of this application, the method includes: receiving search prompts via a large language model, the search prompts including prompts related to a target gene; determining the search target of the search prompts via the large language model, wherein different search targets have a first mapping relationship with different scene agents, each scene agent encapsulating processing logic corresponding to the search target, the processing logic including a main database to be searched; generating a search strategy via a target scene agent having a first mapping relationship with the search target, wherein the search strategy includes search keywords and a target main database, the target main database being the main database in the logic encapsulated by the target scene agent; and searching the target main database according to the search keywords to determine search results related to the target gene sequence.

[0007] The above aspects provide an intelligent interactive automated retrieval method based on a large language model. The front-end scene agent (agent) interacts with the user in multiple rounds using natural language, accurately identifying the user's retrieval target. Furthermore, for each retrieval target, a customized retrieval strategy is automatically generated on the back-end to help users complete bioinformatics retrieval, effectively improving retrieval efficiency.

[0008] According to a first aspect of this application, in some implementations, there is a second mapping relationship between different search targets and different prompt word templates; after determining the search target of the search prompt word through the large language model, the method further includes: obtaining a target prompt word template that has a second mapping relationship with the search target, wherein the target prompt word template includes at least one pair of example information, wherein each pair of example information includes a prior search prompt word with the search target and its corresponding prior search keyword; and determining the search keyword corresponding to the current search prompt word based on the target prompt word template and the current search prompt word.

[0009] According to a first aspect of this application, in some implementations, generating a retrieval strategy through a target scene agent having a first mapping relationship with the retrieval target includes: determining retrieval parameters based on at least one of the following information, wherein the retrieval parameters include retrieval filtering conditions, and the information includes: default parameters associated with the target retrieval scene and received user input parameters; and determining a retrieval strategy based on the retrieval keywords, the target master database, and the retrieval parameters.

[0010] According to a first aspect of this application, in some implementations, when it is determined that the search result does not meet the first requirement, the method further includes: determining the areas for improvement in the search result through the large language model; and determining an optimized search strategy based on the search strategy, the search result, and the areas for improvement.

[0011] According to a first aspect of this application, in some implementations, the improvement point includes at least one of the following: adjusting search parameters, determining auxiliary databases to expand the search scope, optimizing the query route when the search scope includes at least two databases, and optimizing the search keywords; wherein the query route includes the search order of different databases.

[0012] According to a first aspect of this application, in some implementations, the adjustment of the retrieval parameters includes at least one of the following:

[0013] The first auxiliary function agent analyzes the search results and adjusts the search parameters in the search strategy based on their values.

[0014] Based on the search results analyzed by the first auxiliary function agent, additional search parameters are added to the search parameters included in the search strategy.

[0015] The search results are analyzed by the first auxiliary function agent to reduce the number of search parameters included in the search strategy.

[0016] If it is determined that an auxiliary database will be added, the retrieval parameters for retrieving the auxiliary database are determined by the first auxiliary function agent;

[0017] The first auxiliary function agent is either the target scene agent or another agent called by the large language model.

[0018] According to a first aspect of this application, in some implementations, determining the auxiliary database includes: analyzing the search results through a second auxiliary functional agent to determine at least one auxiliary database;

[0019] The optimization of the query route when the search scope includes at least two databases includes: analyzing the search results through a third auxiliary function agent and adjusting the database query order in the search path of the search strategy; wherein, the second auxiliary function agent and the second auxiliary function agent are the target scene agent or other agents called by the large language model.

[0020] According to a first aspect of this application, in some implementations, optimizing the search keywords includes at least one of the following:

[0021] The fourth auxiliary function intelligent agent determines the synonyms of the search keywords in the search strategy;

[0022] The fourth auxiliary intelligent agent is used to modify the search keywords in the search strategy;

[0023] The fourth auxiliary function agent is either the target scene agent or another agent called by the large language model.

[0024] According to a first aspect of this application, in some implementations, retrieving the target master database based on the search keywords to determine search results for the target gene sequence includes: determining a vectorized database corresponding to the target master database, wherein the vectorized database includes multiple document vectors, and the document vectors are obtained by vectorizing the documents in the target master database; determining a search vector corresponding to the search keywords; calculating the vector similarity between the search vector and the document vectors in the vectorized database; and determining search results for the target gene sequence based on the vector similarity, wherein the search results are determined based on document vectors with a vector similarity greater than a preset threshold, or the search results are determined based on the top preset number of document vectors with the highest vector similarity; or...

[0025] Calculate the matching degree between the search keywords and documents in the target master database; determine the search results for the target gene sequence based on the matching degree, wherein the search results are determined based on documents with a matching degree greater than a preset threshold, or the search results are determined based on the top preset number of documents with the highest matching degree.

[0026] According to a first aspect of this application, in some implementations, if the retrieval process is based on non-vector form search keywords and the retrieval results do not meet the second requirement, the method further includes at least one of the following: determining the search vector corresponding to the search keyword, and performing a retrieval in a vectorized database based on the search vector, wherein the vectorized database is obtained by vectorizing the documents in the database to be retrieved; determining synonyms of the search keyword, using the synonyms as search keywords, and performing a retrieval based on the search keywords.

[0027] According to a first aspect of this application, in some implementations, the search results include multiple genomes, and the method further includes: for each genome in the search results, extracting the following information: unique access identifier, assembly level, and assembly quality index; determining the display priority of the genome based on the assembly level and assembly quality index, wherein the display priority of genomes at the complete assembly level is higher than that of genomes at the incomplete assembly level, and the display priority of genomes at the same assembly level is determined according to the assembly quality index; sorting the unique access identifiers of the genomes according to the display priority, and generating feedback results.

[0028] According to a first aspect of this application, in some implementations, there is a third mapping relationship between different search targets and different filtering mechanisms; when the search results meet the third requirement, the method further includes: obtaining a target filtering mechanism that has a third mapping relationship with the search target; filtering the search results based on the target filtering mechanism, wherein the filtered search results are used to provide feedback to the user.

[0029] According to a first aspect of this application, in some implementations, when the search target includes pathogen detection, the search results are filtered based on the target filtering mechanism, including: filtering out a first result set with the component level being a complete genome; if the first result set meets the third requirement, determining a feedback result based on the first result set; if the first result set does not meet the third requirement, filtering out a second result set with the attribute of a reference genome from the first result set, and determining a feedback result based on the second result set.

[0030] According to a first aspect of this application, in some implementations, the target gene sequence is used as a target region to be amplified in a polymerase chain reaction.

[0031] A second aspect of this application proposes a gene information retrieval device. According to an embodiment of this application, the system includes: a receiving module, configured to receive search prompts via a large language model, the search prompts including prompts related to a target gene; a first determining module, configured to determine the search target of the search prompts via the large language model, wherein different search targets have a first mapping relationship with different scene agents, each scene agent encapsulating processing logic corresponding to the search target, the processing logic including a main database to be searched; a generating module, configured to generate a retrieval strategy via a target scene agent having a first mapping relationship with the search target, wherein the retrieval strategy includes search keywords and a target main database, the target main database being the main database in the logic encapsulated by the target scene agent; and a retrieval module, configured to search the target main database based on the search keywords to determine retrieval results related to the target gene sequence.

[0032] The description of the beneficial effects in the first aspect also applies to this aspect, and will not be repeated here.

[0033] In a third aspect, this application provides an electronic device. According to an embodiment of this application, the device includes: a processor and a memory; the memory for storing a computer program; and the processor for executing the computer program to implement the method described in the first aspect.

[0034] In a fourth aspect, this application provides a computer-readable storage medium. According to an embodiment of this application, the computer-readable storage medium includes computer instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect.

[0035] In a fifth aspect, this application provides a computer program product. According to an embodiment of this application, the computer program product includes computer instructions that, when some or all of the computer instructions are executed on a computer, cause the method described in the first aspect to be performed.

[0036] The aforementioned electronic devices, computer-readable storage media, and computer program products achieve highly efficient automation by automatically executing gene information retrieval methods through computer instructions. This reduces the complexity of human intervention in the retrieval process of target gene sequences and provides a more efficient gene sequence retrieval scheme.

[0037] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of a gene information retrieval method provided in one embodiment of this application;

[0040] Figure 2 This is a schematic diagram of a multi-agent retrieval planning architecture provided in one embodiment of this application;

[0041] Figure 3 This is a schematic diagram of the databases corresponding to different scenarios in the online API retrieval process in one embodiment of this application;

[0042] Figure 4 This is a schematic diagram of the vector retrieval process in one embodiment of this application;

[0043] Figure 5 This is a schematic diagram of the Self-ReAct inference feedback mechanism in one embodiment of this application;

[0044] Figure 6 This is a flowchart illustrating the optimization of the retrieval strategy in one embodiment of this application;

[0045] Figure 7 This is a schematic diagram of a gene information retrieval device provided in one embodiment of this application;

[0046] Figure 8 This is a schematic diagram of an electronic device provided in one embodiment of this application. Detailed Implementation

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

[0048] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0049] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0050] Multiplex polymerase chain reaction (mPCR) is a widely used technique in molecular biology that amplifies more than one specific DNA fragment in vitro using a pair or more primers. It is widely applied in research such as gene diagnosis, tumor diagnosis, pathogen detection, genotyping, and targeted capture sequencing. Multiplex PCR requires the specific amplification of multiple regions within the same PCR reaction system, and all amplification primers must cooperate to achieve good amplification results. When designing primers for multiplex PCR experiments, the target region to be amplified must first be determined. If using gene sequence retrieval schemes provided by relevant technologies, users need to manually select relevant databases, enter search keywords, and complete the data retrieval through multiple clicks or script execution. Advanced users or researchers write and run scripts, interacting with the database through a command-line interface to perform tasks such as data filtering, comparison, and annotation. Users accessing public databases via web interfaces need to manually select relevant options, enter keywords, and complete the data retrieval through multiple clicks.

[0051] It is evident that the aforementioned technologies have the following drawbacks in application:

[0052] 1. Using web interfaces or command-line scripts for bioinformatics retrieval requires users to possess advanced professional knowledge and technical skills, especially in keyword setting, command-line operation, and script writing. This presents a high learning and usage cost for non-professional users or novice researchers, limiting the rapid retrieval, acquisition, and application of bioinformatics data.

[0053] 2. The operation process is complex. Users need to manually select relevant databases, set filter conditions, and perform multiple rounds of repetitive searching and manual information filtering to obtain key information or data. This process is not only repetitive and tedious, but also prone to errors and missed information, reducing work efficiency.

[0054] 3. Traditional search tools typically only provide search results and lack the ability to interpret and analyze them. Users find it difficult to obtain advice on search strategies or gain a deeper understanding of the results during the search process, which limits the depth and breadth of research.

[0055] 4. For different detection needs, such as protein mutation detection, genetic disease detection, cancer detection and pathogen detection, traditional methods require users to adjust and filter the retrieved data multiple times, which further increases the complexity and inconvenience of the retrieval process.

[0056] To address the aforementioned shortcomings, the inventors provide an intelligent, interactive, and automated retrieval scheme based on a large language model. The front-end intelligent agent interacts with the user in multiple rounds using natural language to accurately identify the user's search targets. Furthermore, for each search target, a customized search strategy is automatically generated in the back-end to assist the user in completing bioinformatics retrieval. Specifically, the retrieval scheme provided in this application can be used for downstream multiplex PCR primer design and experimental testing, but is not limited to this.

[0057] In one aspect of this application, a method for retrieving gene information is proposed, referring to... Figure 1 The method includes:

[0058] S110, receive search suggestions through a large language model, the search suggestions including suggestions related to the target gene.

[0059] The information related to the target gene mentioned above can include gene sequence information, drug resistance mutation sites, gene disease information, gene annotations, etc. Therefore, the search suggestions could be keywords related to the target gene sequence, drug resistance mutation sites, gene diseases, or gene annotations, etc.

[0060] In this paper, the gene information retrieval system provides a user interaction module, which specifically receives user-input search suggestions through an integrated large language model (LLM). These search suggestions inform the system of the user's current search target or direction. An example search suggestion is "find information on gene variants related to breast cancer." Users input these search suggestions in natural language, eliminating the need for programming languages ​​or skills (e.g., command-line tools), effectively lowering the technical barrier.

[0061] Example search suggestions could be: "You are an AI assistant designed to help users retrieve biometric information sequences. When a user requests protein mutation detection, you should call the `protein_mutation_search` function to obtain protein and gene information; furthermore, you should use the same language the user uses when replying to the user." Another example search suggestion could be: "You are an AI assistant designed to help users retrieve biometric information sequences. When a user requests cancer analysis or wants to obtain cancer-related genes, you should call the `cancer_search` function to obtain the target gene sequence and SNP information. Furthermore, you should use the same language the user uses when replying to the user."

[0062] In this paper, the gene information retrieval system provides a retrieval planning module, which can specifically determine the user's retrieval target and the corresponding retrieval strategy through the aforementioned large language model. The specific implementation method for determining the retrieval target can be referred to the embodiment corresponding to S120, and the specific implementation method for determining the retrieval strategy corresponding to the retrieval target can be referred to the embodiment corresponding to S130.

[0063] S120, the retrieval target of the retrieval prompt is determined through the large language model, wherein different retrieval targets have a first mapping relationship with different scene agents, and each scene agent encapsulates the processing logic corresponding to the retrieval target, the processing logic including the main database to be searched.

[0064] In this paper, the aforementioned large language model is a trained model, which can be used to identify the user's search target. Specifically, a set of training samples includes search keywords and their corresponding search targets. Thus, based on the search suggestions received in S110, the user's search target (referred to as the search target for ease of explanation) can be understood and parsed.

[0065] Different search targets correspond to different search scenarios. For example, if the search target is to search for gene sequences related to cancer, it corresponds to a cancer detection scenario; if the search target is to search for gene sequences related to hereditary diseases, it corresponds to a hereditary disease detection scenario; the search target can also be to search for gene sequences related to enzyme engineering (proteins) or gene sequences related to pathogen detection, etc., corresponding to enzyme engineering (protein) detection scenarios and pathogen detection scenarios, respectively.

[0066] In some embodiments, if the initial search suggestions input by the user are ambiguous or incomplete, the aforementioned large language model can engage in multi-turn dialogues with the user through the user interaction module to ensure the accuracy of the identified search target. Since different scene agents (i.e., agent proxies) are provided for different search targets in this embodiment, and each scene agent encapsulates the processing logic corresponding to the search target, ensuring the accuracy of the search target determination is beneficial for providing targeted search strategies, thereby ensuring the accuracy of the search results.

[0067] For example, Figure 2 This is a schematic diagram of a multi-agent retrieval planning architecture provided in one embodiment of this application.

[0068] refer to Figure 2 The large language model (as shown in the master LLM 201 in the figure) performs intent recognition on the received input information. Further, refer to... Figure 2The target detection task can be assigned to the corresponding scene agent by the retrieval agent 202. Since each scene agent encapsulates the processing logic for the corresponding retrieval target, the scene agent then specifically completes the retrieval task for that scene. (Reference) Figure 2 The scene-based intelligent agents include: protein mutation detection scene-based intelligent agent 203, genetic disease detection scene-based intelligent agent 204, cancer detection scene-based intelligent agent 205, pathogen detection scene-based intelligent agent 206, etc. Each scene-based intelligent agent executes encapsulated processing logic, which may involve information interaction with the user during processing to ensure that the search results meet the user's needs. Therefore, in the gene sequence retrieval process provided in this application embodiment, the user can engage in multiple rounds of natural language interaction with different scene-based intelligent agents. The scene-based intelligent agents complete complex retrieval tasks based on the interaction, without requiring the user to possess advanced professional knowledge or programming skills. This allows even non-professional users and beginners to easily use the system, reducing the barrier to entry and time costs. Furthermore, through information from multiple rounds of interaction with the user, the scene-based intelligent agents can quickly understand user needs and automatically select appropriate databases and retrieval strategies, reducing the time and steps required for manual operation. The dynamically optimized retrieval strategies of the scene-based intelligent agents ensure that they can follow the dynamic changes in user needs, improving the efficiency of data retrieval and significantly shortening the time for users to obtain usable data.

[0069] In this embodiment, by setting up intelligent agents for different scenarios, customized retrieval strategies are provided for different PCR experimental information retrieval scenarios, which helps to meet diverse research needs. Therefore, whether it is pathogen detection, protein mutants, or genetic disease detection, the retrieval system provided by this application can provide professional solutions, greatly improving the flexibility and applicability of PCR experiments.

[0070] Continue to refer to Figure 1 S130, a retrieval strategy is generated by a target scene agent that has a first mapping relationship with the retrieval target, wherein the retrieval strategy includes retrieval keywords and a target master database, and the target master database is the master database in the logic encapsulated by the target scene agent.

[0071] For example, the above-mentioned large language models (such as...) Figure 2The master control LLM (Large Language Model) generates formatted instructions to call target functions, thereby activating the target scenario agent. For example, if a user inputs a search suggestion in natural language as "find breast cancer-related gene mutations," the large language model parses the search suggestion to determine the search target as: searching for cancer-related gene fragments, i.e., a cancer detection scenario. Further, it can trigger function call instructions, such as `cancer_search({"term":"breast cancer"})`. The large language model can output structured instructions, which may include the search target (e.g., cancer detection), the function name (e.g., `cancer_search`), and search-related parameters (e.g., `{"term":"breast cancer"}`). The target function described above is used as an example to activate the agent in a tumor detection scenario:

[0072]

[0073] This function is used to retrieve the sequence and single nucleotide polymorphism (SNP) information of a target gene in a cancer analysis scenario using multiplex PCR experiments. In this embodiment, after the large language model determines the above-mentioned retrieval target, the corresponding scenario agent can be activated by calling the above function. Through the processing logic encapsulated by the agent, the sequence and SNP data of a specific gene for cancer research (such as gene mutation analysis, primer design, etc.) can be obtained. Specifically, the above-mentioned processing logic includes automatically retrieving the following information from the corresponding biological database based on cancer-related terms (i.e., search keywords, such as cancer name, gene name, etc.): the target gene sequence (such as the DNA sequence of cancer-related genes) and SNP information, i.e., cancer-related single nucleotide polymorphism (SNP) site data (such as mutation location, clinical significance). For example, the corresponding biological database can be a database bound in the logic encapsulated by the target scenario agent.

[0074] For example, the target scene intelligent agent receives the structured instructions, generates a retrieval strategy based on the received information, and executes the encapsulated processing logic based on the retrieval strategy, thereby automatically achieving retrieval for the current retrieval target. When the user and the large language model involve multi-turn dialogues, the target scene intelligent agent, in determining the retrieval strategy, not only refers to the information contained in the structured instructions but also to the multi-turn dialogue history (such as the user-added "malignant subtype") to improve the completeness and precision of the retrieval strategy.

[0075] In an exemplary embodiment, the retrieval strategy includes retrieval keywords, a database to be retrieved, and may also include retrieval parameters. The database to be retrieved during the initial retrieval process may be the main database bound to the processing logic encapsulated by the current scene agent. For example, in a cancer detection scenario, the database bound to the scene agent is a database specifically containing cancer-related genomes, such as the Cosmic database. Therefore, in this embodiment, the scene agent can specifically and automatically determine the most suitable database for the current scenario without requiring the user to understand the database structure, query language (such as SQL), or operation of specific software (such as Cosmic, BLAST, GATK, etc.). Thus, even researchers without a bioinformatics background can quickly and efficiently retrieve relevant gene fragments, significantly lowering the technical threshold.

[0076] The following embodiments describe how to determine search keywords. Since search keywords play a crucial role in the retrieval process, this application embodiment also provides the following scheme to improve the accuracy of search keyword determination. Specifically, there is a second mapping relationship between different search targets and different prompt word templates. Because the requirements for extracting search keywords differ in different scenarios—for example, pathogen detection requires handling ambiguous species names, cancer detection requires associating gene variation terms, etc.—this application embodiment improves the accuracy of search keywords in various scenarios through scenario customization.

[0077] Specifically, after determining the search target of the above search suggestions, the following steps are performed to extract search keywords.

[0078] In S11, a target prompt word template that has a second mapping relationship with the above-mentioned search target is obtained.

[0079] Each prompt term template includes at least one pair of example information. For example, any pair of example information in the target prompt term template above includes: a prior search prompt term with the search target and its corresponding prior search keyword. Referring to the prompt term template shown in Table 1, it includes three example information; regarding the first example, its search prompt term is "Species identification of Salmonella enterica subspecies enterica serovar Ugandavarieties 15+", and its corresponding prompt term is ""species_information":["Salmonella enterica subspecies enterica serovar Ugandavarieties 15+"]. In this embodiment, few-shot prompt term templates are provided for various scenarios to provide correct standard examples for the extraction of search keywords in the current scenario. By guiding the large language model to understand the search target and data format of a specific scenario through examples, the accuracy and efficiency of keyword extraction can be effectively improved, thereby ensuring the accuracy of the search results.

[0080] Table 1

[0081]

[0082]

[0083] In S12, based on the target prompt term template and the current search prompt term, the search keywords corresponding to the current search prompt term are determined.

[0084] By using example data from the target prompt template as prompts, the large language model is guided to understand the retrieval task of the current retrieval target. Thus, the examples in the target prompt template guide the large language model to generate outputs relevant to the current retrieval target, thereby optimizing the retrieval process. Assume there are K pairs of example data in the target prompt template: (x1, y1), (x2, y2), ..., (x... K ,y K Formula (1) can be used to determine how to determine the current search suggestion term x. K+1 Corresponding search keywords

[0085]

[0086] Where p represents the K pairs of sample data containing the target prompt word template, and K is a positive integer.

[0087] Taking cancer detection as an example, the user's input search keywords could be "find EGFR gene mutations related to lung cancer". The search scenario of obtaining cancer-related gene sequences has a second mapping relationship with the cancer detection prompt word template. Furthermore, the big language model understands the search target and data format of the cancer scenario based on the example in the template, and can output the corresponding search keywords such as: {"Disease type":"Lung cancer","Gene name":"EGFR","Target information":"Mutation"}.

[0088] Taking pathogen detection as an example, the user's input search keyword could be "obtain the whole genome sequence of the novel coronavirus". The search scenario of obtaining pathogen-related gene sequences has a second mapping relationship with the pathogen detection prompt word template. Furthermore, the large language model understands the search target and data format of the pathogen detection scenario based on the example in the template, and can output the corresponding search keywords such as: {"pathogen name":"SARS-CoV-2","genome characteristics":"whole genome"}.

[0089] As can be seen, the prompt word templates in this embodiment are strongly correlated with the scene, using independent prompt word templates for different retrieval scenarios to adapt to their terminology features and retrieval targets, while avoiding semantic biases of general templates and ensuring the accuracy of terminology extraction. Through scene-specific input-output examples, the large language model is guided to accurately extract keywords and classify scenes, significantly improving the accuracy of intent recognition. Furthermore, when adding a new scene, only the corresponding Few-Shot template needs to be designed, without global adjustments, thus facilitating dynamic expansion.

[0090] In an exemplary embodiment, search parameters, such as search filtering conditions, can be determined based on default parameters associated with the target search scenario and / or received user input parameters. Furthermore, the search strategy for the current search can be determined based on the search keywords, the target master database, and the aforementioned search parameters determined in the above embodiments.

[0091] Continue to refer to Figure 1 S140, Search the target master database according to the search keywords to determine the search results for the target gene sequence.

[0092] Following the implementation method corresponding to S130, after the target scenario agent has initially determined the above-mentioned retrieval strategy, including retrieval keywords and the main database in the logic encapsulated by the current agent, the target main database can be searched according to the retrieval keywords to determine the retrieval results regarding the above-mentioned target gene sequence.

[0093] The gene sequence retrieval system provided in this application embodiment also includes a data retrieval module. Specifically, the data retrieval module interfaces with multiple authoritative biological databases to perform actual data retrieval operations. The data retrieval module receives retrieval requests from scene agents. Since the main databases encapsulated in the logic of different scene agents may differ, the module can intelligently schedule the application programmer interfaces (APIs) of relevant databases according to the retrieval requests to achieve the retrieval. Figure 3 This is a schematic diagram illustrating the databases corresponding to multiple scenarios in the online API retrieval process of one embodiment of this application. Combined with... Figure 3 The embodiments of this application can perform the following steps to achieve online API retrieval.

[0094] In S21, the matching degree between the search keywords and documents in the target master database is calculated.

[0095] By using a keyword matching algorithm to calculate the matching degree between the search keywords and documents in the target main database, the relevance of the document to the query can be determined, thereby achieving accurate retrieval. For example, given the search keyword q and the document d in the database, the matching degree can be calculated using formula (2).

[0096] rel(q,d)=∑ k∈q match(k,d) (2)

[0097] Here, the search keyword q is a set of keywords, for example, the search keywords for the search suggestion "breast cancer gene mutation" are: ["breast cancer", "gene", "mutation"]; k represents a single keyword in the search keyword set q, such as "breast cancer". The document d in the database is the target data entry to be matched. For example, a document in the database contains the following fields: "gene name":"BRCA1", "disease type":"breast cancer", "mutation type":"missense mutation".

[0098] The `match(k,d)` function represents the degree of match between keyword `k` and document `d`. Specifically, it can be a binary function, where a match is assigned a value of 1, and a non-match is assigned a value of 0. For example, if `k` = "breast cancer" and the "disease type" field of `d` is "breast cancer", then `match(k,d)` = 1.

[0099] rel(q,d) represents the relevance score between q and d, which is the sum of the matching scores of all keywords in q. For example, if q = ["breast cancer", "gene", "mutation"], and d matches "breast cancer" and "mutation", then rel(q,d) = 1 + 0 + 1 = 2.

[0100] In S22, based on the matching degree mentioned above, the search results for the target gene sequence are determined.

[0101] For example, the search results can be determined based on documents with a matching degree greater than a preset threshold. Alternatively, the search results can be determined based on the top preset number of documents with the highest matching degree. This allows the user to be provided with the gene sequences in the database that best match the search keywords.

[0102] The above-described retrieval methods are applicable when each retrieval keyword (i.e., bioinformatics entity) is accurately extracted, enabling precise queries and efficient retrieval. In addition to the above retrieval methods, this application embodiment also provides a vector database and performs retrieval based on vectors. This reduces the possibility of invalid retrieval due to inaccurate keyword extraction. For example, if the extracted retrieval keyword is "Severe acute respiratory syndrome 2," but lacks the "coronavirus" component from the standard bioinformatics entity "Severe acute respiratory syndrome coronavirus 2," the above retrieval methods will not yield valid results. Therefore, this application, by deploying vector retrieval, facilitates obtaining more comprehensive retrieval results. Specifically, this application embodiment can perform the following steps to achieve vector retrieval.

[0103] In S31, the vectorized database corresponding to the aforementioned target master database is determined.

[0104] The vectorized database includes multiple document vectors, which are obtained by vectorizing each document in the target master database. (Reference) Figure 4 The documents in the original document database are vectorized to obtain the corresponding vectorized database.

[0105] In S32, the search vector corresponding to the search keyword is determined.

[0106] refer to Figure 4 A vector space model is used to map search keywords to documents in the target database into the same vector space, resulting in the aforementioned search vector and document vector. High-dimensional vector representation enables embedded storage of bioinformatics data, thereby efficiently calculating the similarity between queries and documents.

[0107] In S33, the vector similarity between the retrieval vector and the document vector in the vectorized database is calculated.

[0108] For example, the similarity of the above vectors can be determined by formula (3).

[0109]

[0110] Where q represents the retrieval vector, d i This represents the i-th document vector in the vectorized database, where i is a positive integer.

[0111] In S34, the retrieval results for the target gene sequence are determined based on the vector similarity.

[0112] The search results are determined based on document vectors with a similarity greater than a preset threshold. For example, the preset threshold for similarity can be set to 0.8, and search results with similarities greater than this threshold will be returned after similarity calculation. Alternatively, the search results can be determined based on the top preset number of document vectors with the highest similarity. For example, the top 10 are selected and returned based on descending similarity. For instance, the search results are presented to the user in a table format on the front end, allowing English-speaking users to select and download the corresponding sequence information. This search mechanism ensures fast and accurate retrieval of the biological data most relevant to the user's query.

[0113] It is understood that one or both of the two search methods can be used for searching according to actual needs, and this application embodiment does not limit this.

[0114] It should be noted that if the retrieval process is based on non-vector form search keywords, and the search results do not meet the second requirement (such as too few or empty search results), the embodiments of this application also provide the following methods.

[0115] Method 1

[0116] The process involves determining the retrieval vector corresponding to the search keywords and then performing a search in a vectorized database based on these retrieval vectors. This vector-based retrieval method reduces the possibility of invalid searches due to inaccurate keyword extraction. For example, the extracted keyword "Severe acute respiratory syndrome 2" lacks the "coronavirus" component compared to the standard bioinformatics entity "Severe acute respiratory syndrome coronavirus 2".

[0117] Method 2

[0118] The process involves identifying synonyms for the search keywords and using these synonyms as the search keywords themselves. For example, if a user inputs "Whole Genome Sequencing of SCoV2", this embodiment extracts "SCoV2" as the search keyword. If an online API database search is first performed, but the NCBI database does not contain the name "SCoV2", the search results will be empty (i.e., the second requirement mentioned above is not met). In this case, a synonym conversion method can be activated to convert "SCoV2" into the more common name "Severe acute respiratory syndrome coronavirus 2". Then, the search process can be performed again based on the synonyms to obtain the corresponding search results.

[0119] In an exemplary embodiment, this application provides a Self-ReAct inference feedback mechanism to perform self-checking of search results. Specifically, the Self-ReAct inference feedback mechanism ensures the accuracy and comprehensiveness of search results by automatically adjusting query parameters and integrating results from multiple databases, thereby enabling the system to cover more data sources and provide more detailed and reliable sequence information. For example, refer to... Figure 5 In the t-th round of the search process, which is executed based on the t-th round strategy, after obtaining the t-th round search results, it is determined whether the first requirement mentioned above is met (e.g., in a PCR scenario, the target region to be amplified cannot be clearly identified through the search). If the first requirement is not met, the t-th round search strategy is optimized, and the search process is executed again after obtaining the t+1-th round search strategy. Specifically, if it is determined that the current search results are insufficient to meet the user's needs (i.e., the first requirement mentioned above), it also provides, for example... Figure 6 The proposed solution optimizes the retrieval strategy.

[0120] refer to Figure 6 In S210, the areas for improvement in the current search results are identified using a large language model. In S220, an optimized search strategy is determined based on the current search strategy, the current search results, and the aforementioned areas for improvement.

[0121] For example, the retrieval strategy can be optimized using the Update function, as shown in formula (4).

[0122] p t+1 =Update(p t ,y t ,f t (4)

[0123] Where, p t Let y represent the retrieval strategy for the t-th round of retrieval. tf represents the search results after the search strategy in the t-th round of retrieval. t This represents a reflection on the results of the t-th round of retrieval, i.e., areas for improvement.

[0124] Among them, the points to be improved include at least one of the following:

[0125] (1). Adjust the search parameters.

[0126] (2) Identify auxiliary databases to expand the search scope.

[0127] (3) Optimize the query route when the search scope includes at least two databases; wherein the above query route includes the search order of different databases.

[0128] (4) Optimize the search keywords.

[0129] Regarding "(1). Adjusting search parameters", it can be achieved through at least one of the following adjustment methods:

[0130] (11). The first auxiliary function agent analyzes the search results after the t-th round of retrieval and adjusts the search parameter values ​​in the t-th round of retrieval strategy. For example, if the search parameter s in the t-th round of retrieval strategy is greater than 0.8, the adjusted search parameter s is greater than 0.85.

[0131] (12). Analyze the retrieval results of the tth round through the first auxiliary function agent, and add other retrieval parameters based on the retrieval parameters included in the tth round retrieval strategy; for example, if the retrieval results are too broad, it is necessary to add other retrieval parameters to make the retrieval results more targeted and accurate.

[0132] (13). Analyze the retrieval results of the tth round through the first auxiliary function agent and reduce the retrieval parameters included in the tth round retrieval strategy; if there are too few retrieval results, it is necessary to reduce one or more retrieval parameters to make the retrieval results more reasonable.

[0133] (14). If it is determined that an auxiliary database is to be added, the retrieval parameters for retrieving the added auxiliary database are determined by the first auxiliary function agent.

[0134] Among them, the first auxiliary function agent is the aforementioned target scenario agent or other agents called by the large language model.

[0135] Regarding "(2). Determining the auxiliary database", this application provides the following adjustment method in its embodiments:

[0136] (21) Analyze the retrieval results using a second auxiliary function agent to determine at least one auxiliary database;

[0137] Regarding "(3). Optimize query route", this application provides the following adjustment method:

[0138] (31) Analyze the retrieval results through the third auxiliary function agent and adjust the database query order in the retrieval path of the retrieval strategy; wherein, the second auxiliary function agent and the third auxiliary function agent are target scene agents or other agents called by the large language model.

[0139] Regarding "(4). Optimize the search keywords", this application provides the following adjustment methods:

[0140] (41) Using the fourth auxiliary function intelligent agent, determine the synonyms of the search keywords in the search strategy;

[0141] (42) The search keywords in the search strategy are modified by the fourth auxiliary function intelligent agent;

[0142] The fourth auxiliary function agent is either the target scene agent or another agent called by the large language model.

[0143] A specific example of the ReAct inference feedback mechanism is as follows: For the search suggestion "find the whole genome sequence of the novel coronavirus (SCoV2)", if the t-th round search strategy (initial request) p t The keyword is SCoV2, the database is NCBI Genome (online API), and the search parameter is assembly_level = Complete Genome. The search results y under the t-th round search strategy are... t No matching results found (because the standard name in NCBI is "SARS-CoV-2").

[0144] Determine the retrieval result y under the t-th round retrieval strategy using a large language model. t Existing areas for improvement f t "SCoV2 is a vague abbreviation and needs to be converted to the standard term SARS-CoV-2."

[0145] The adjustment strategies determined by the Update function include: replacing SCoV2 with SARS-CoV-2 through keyword optimization; maintaining the search parameter assembly_level = Complete Genome; and calling the local vector library for fuzzy matching since NCBI still did not find the match.

[0146] Therefore, the retrieval strategy (optimized retrieval request) for the (t+1)th round is determined by the Update function. t+1The keyword was SARS-CoV-2, the database was NCBI Genome+ local vector library, and the search parameter was assembly_level=CompleteGenome.

[0147] Furthermore, the above-mentioned retrieval strategy p in round t+1 is executed. t+1 This yields the search results for round t+1: standard genomes retrieved from NCBI, and similar sequences supplemented from the local vector library.

[0148] In the embodiments described above, the large language model analyzes the shortcomings of the current search results using the Self-ReAct inference mechanism, adjusts query parameters or expands the search scope, formulates a revised and optimized search strategy, and re-initiates the search request. For example, in a PCR scenario, if the target region to be amplified cannot be clearly identified through the search, the above mechanism is used to adjust the query strategy, re-search and integrate results from multiple databases to improve data accuracy and coverage. Through real-time search feedback, the gene sequence retrieval system can self-adjust its search content, enhancing its adaptability to biological data.

[0149] After determining the search results through the above embodiments, the search results can be further processed to provide a brief summary to the user. For example, the search results include multiple genomes, and this embodiment also performs the following steps to determine the feedback results.

[0150] In S41, for each genome in the above search results, the following information is extracted: unique access identifier, assembly level, and assembly quality index.

[0151] Each genome has a unique access identifier (e.g., GCF_000001405.40), which can be used to accurately locate the genome in a database (e.g., NCBI).

[0152] The assembly level of a genome indicates the completeness of the genome assembly. Specifically, it can include: Complete Genome (no gaps), Scaffold (with gaps and connections), and Contig (completely contiguous). The assembly level can affect the reliability of subsequent analyses. For example, genomes can be classified as complete or incomplete based on the assembly level value. A complete genome is a contiguous sequence without gaps, typically corresponding to an NCBI GCF number; a noncomplete genome consists of scaffolds with gaps or isolated contigs, typically corresponding to a GCA number.

[0153] Assembly quality metrics include Contig N50, which reflects the assembly quality of undamaged continuous sequences, and Scaffold N50, which reflects the assembly quality of scaffold sequences with notches. Contig N50 represents the contig length when contigs are sorted by length and summed to 50% of the total length; it reflects the splicing capability of short-read sequencing and can be used to assess basic splicing reliability. Scaffold N50 represents the scaffold length when scaffolds are sorted by length and summed to 50% of the total length; it reflects the scaffold construction capability of long-read or paired-end data and can be used to assess the integrity of large fragment assembly.

[0154] It is understandable that the information extracted from each genome is not limited to the above. Other metadata can also be extracted according to actual needs, such as sequencing technology (e.g., PacBio HiFi, Illumina), species classification (to avoid sample contamination or mislabeling), etc.

[0155] In S42, the display priority of each genome is determined based on its assembly level and assembly quality indicators. And in S43, the unique access identifiers of the genomes are sorted according to the display priority, generating feedback results.

[0156] Among them, the display priority of genomes at the complete assembly level is higher than that of genomes at the incomplete assembly level, and the display priority of genomes at the same assembly level is determined according to the assembly quality index.

[0157] For example, for complete genomes (GCF type genomes), the display priority is positively correlated with the scaffold N50 value, meaning they are sorted in descending order of scaffold N50 value, allowing users to prioritize genomes with the highest scaffold N50 value. For instance, if genome A is complete with scaffold N50 = 50Mb and genome B is complete with scaffold N50 = 30Mb, then the display priority is A > B. For incomplete genomes (GCA type genomes), the display priority is positively correlated with the contig N50 value, meaning they are sorted in descending order of contig N50 value, allowing users to select the genome with the highest contig N50 value as the best candidate.

[0158] Regarding the search results provided to users, this embodiment also offers enhanced interpretation functionality. For example, in addition to displaying basic sequence information, it provides detailed information such as gene function descriptions, gene locations, and SNP sites to help users better understand and apply the search results. Users can confirm or adjust the results as needed to further improve the accuracy and reliability of the data.

[0159] In an exemplary embodiment, when the feedback results contain a large amount of content (i.e., meet the third requirement), such as when the system retrieves data and there are hundreds of results (because, for example, there may be multiple strains of the same species with different sequences), a filtering mechanism is provided to facilitate user viewing and understanding. Specifically, different filtering mechanisms can be set for different search scenarios / search targets to achieve precise filtering based on the scenario. For example, the following steps can be performed to filter the search results.

[0160] In S51, a target filtering mechanism that has a third mapping relationship with the above-mentioned search target is obtained.

[0161] There is a third mapping relationship between different search targets and different filtering mechanisms.

[0162] In S52, the search results are filtered based on the target filtering mechanism, and the filtered search results are used to provide feedback to the user.

[0163] When the search target includes pathogen detection, the specific implementation of filtering the search results based on the target filtering mechanism includes:

[0164] In S521, the first set of results is selected at the component level as a complete genome.

[0165] In S522, if the first result set satisfies the third requirement, a feedback result is determined based on the first result set.

[0166] In S523, if the first result set does not meet the third requirement, a second result set with the attribute of reference genome is selected from the first result set, and a feedback result is determined based on the second result set.

[0167] For genomes retrieved in pathogen detection scenarios, if the search results contain a large amount of data, their assembly level is checked first. Genomes with an assembly level of "Complete Genome" are retained to obtain the first result set.

[0168] If the first result set meets the third requirement mentioned above, then the feedback result is determined based on the first result set; specifically, the genomes in the first result set are sorted in descending order according to the scaffold N50 value, so that users can preferentially select the genome with the largest scaffold N50 value.

[0169] If the first result set does not meet the third requirement mentioned above, further filtering is required. Specifically, the representative attribute of the first result set is further examined. Entities in the first result set with the attribute "reference genome" are designated as the second result set.

[0170] For example, genomes from the second result set are preferentially added to the recommended genome list. For instance, if the second result set meets the third requirement mentioned above, then the second result set is used as the feedback result. If the number of genomes in the second result set is small, the feedback result can be determined jointly based on genomes from the first result set other than those in the second result set.

[0171] For example, if there is no "reference genome" in the first result set, the first result set can be used as user feedback data, or the genomes in the first result set can be sorted in descending order according to the scaffold N50 value, so that users can prioritize the genome with the largest scaffold N50 value.

[0172] As can be seen, by processing the search results as described above (such as extracting key information and genome sequencing, and filtering mechanisms corresponding to different scenarios), a user-friendly interface and a concise result presentation method can be provided, which is conducive to improving the overall search experience of users.

[0173] This application also proposes a genetic information retrieval device. (Reference) Figure 7 The device 700 includes: a receiving module 710, a first determining module 720, a generating module 730, and a retrieving module 740.

[0174] The system includes a receiving module 710, which receives search suggestions via a large language model, the search suggestions including suggestions related to the target gene; a first determining module 720, which determines the search target of the search suggestions via the large language model, wherein different search targets have a first mapping relationship with different scene agents, and each scene agent encapsulates processing logic corresponding to the search target, the processing logic including a main database to be searched; a generating module 730, which generates a search strategy through a target scene agent with a first mapping relationship to the search target, wherein the search strategy includes search keywords and a target main database, the target main database being the main database in the logic encapsulated by the target scene agent; and a searching module 740, which searches the target main database based on the search keywords to determine search results related to the target gene sequence.

[0175] In some examples of this application, there is a second mapping relationship between different search targets and different prompt word templates; the device further includes: an acquisition module and a second determination module; the acquisition module is used to: after the determination module 720 determines the search target of the search prompt word through the large language model, acquire a target prompt word template that has a second mapping relationship with the search target, wherein the target prompt word template includes at least one pair of example information, wherein each pair of example information includes a prior search prompt word with the search target and its corresponding prior search keyword; the second determination module is used to determine the search keyword corresponding to the current search prompt word based on the target prompt word template and the current search prompt word.

[0176] In some examples of this application, the generation module 730 is specifically used to: determine search parameters based on at least one of the following information, wherein the search parameters include search filtering conditions, and the information includes: default parameters associated with the target search scenario and received user input parameters; and determine a search strategy based on the search keywords, the target master database, and the search parameters.

[0177] In some examples of this application, the apparatus further includes: an optimization module; the optimization module is configured to, when it is determined that the search result does not meet the first requirement, determine the areas for improvement of the search result through the large language model; and, based on the search strategy, the search result, and the areas for improvement, determine an optimized search strategy.

[0178] In some examples of this application, the points to be improved include at least one of the following: adjusting search parameters, determining auxiliary databases to expand the search scope, optimizing the query route when the search scope includes at least two databases, and optimizing the search keywords; wherein the query route includes the search order of different databases.

[0179] In some examples of this application, the optimization module is used to adjust retrieval parameters, specifically for at least one of the following:

[0180] The first auxiliary function agent analyzes the search results and adjusts the search parameters in the search strategy based on their values.

[0181] Based on the search results analyzed by the first auxiliary function agent, additional search parameters are added to the search parameters included in the search strategy.

[0182] The search results are analyzed by the first auxiliary function agent to reduce the number of search parameters included in the search strategy.

[0183] If it is determined that an auxiliary database will be added, the retrieval parameters for retrieving the auxiliary database are determined by the first auxiliary function agent;

[0184] The first auxiliary function agent is either the target scene agent or another agent called by the large language model.

[0185] In some examples of this application, when the optimization module is used to determine an auxiliary database, it is specifically used to: analyze the retrieval results through a second auxiliary function agent to determine at least one auxiliary database;

[0186] The optimization module is used to optimize the query route when the search range includes at least two databases. Specifically, it is used to: analyze the search results through a third auxiliary function agent and adjust the database query order in the search path of the search strategy; wherein, the second auxiliary function agent and the second auxiliary function agent are the target scene agent or other agents called by the large language model.

[0187] In some examples of this application, when the optimization module is used to optimize the search keywords, it is specifically used for at least one of the following:

[0188] The fourth auxiliary function intelligent agent determines the synonyms of the search keywords in the search strategy;

[0189] The fourth auxiliary intelligent agent is used to modify the search keywords in the search strategy;

[0190] The fourth auxiliary function agent is either the target scene agent or another agent called by the large language model.

[0191] In some examples of this application, the retrieval module 740 is specifically used for:

[0192] The process involves: determining the vectorized database corresponding to the target master database, wherein the vectorized database includes multiple document vectors, and the document vectors are obtained by vectorizing the documents in the target master database; determining the retrieval vector corresponding to the retrieval keyword; calculating the vector similarity between the retrieval vector and the document vectors in the vectorized database; and determining the retrieval results for the target gene sequence based on the vector similarity, wherein the retrieval results are determined based on document vectors with a vector similarity greater than a preset threshold, or the retrieval results are determined based on the top preset number of document vectors with the highest vector similarity.

[0193] Alternatively, it can be used to: calculate the matching degree between the search keywords and documents in the target master database; and determine the search results for the target gene sequence based on the matching degree, wherein the search results are determined based on documents with a matching degree greater than a preset threshold, or the search results are determined based on the top preset number of documents with the highest matching degree.

[0194] In some examples of this application, if the retrieval process is based on non-vector form search keywords and the retrieval results do not meet the second requirement, the apparatus further includes a fourth determining module; the fourth determining module is used for at least one of the following:

[0195] Determine the retrieval vector corresponding to the retrieval keyword, and perform a retrieval in a vectorized database based on the retrieval vector, wherein the vectorized database is obtained by vectorizing the documents in the database to be retrieved;

[0196] Identify synonyms of the search keywords, use the synonyms as search keywords, and perform a search based on the search keywords.

[0197] In some examples of this application, the search results include multiple genomes, and the apparatus further includes a processing module; the processing module is configured to: for each genome in the search results, extract the following information: unique access identifier, assembly level, and assembly quality index; determine the display priority of the genome according to the assembly level and assembly quality index, wherein the display priority of genomes at the complete assembly level is higher than that of genomes at the incomplete assembly level, and the display priority of genomes at the same assembly level is determined according to the assembly quality index; and sort the unique access identifiers of the genomes according to the display priority to generate feedback results.

[0198] In some examples of this application, the apparatus further includes: a filtering module; a third mapping relationship between different search targets and different filtering mechanisms; the filtering module is used to: obtain a target filtering mechanism that has a third mapping relationship with the search target when the search results meet the third requirement; and filter the search results based on the target filtering mechanism, wherein the filtered search results are used to provide feedback to the user.

[0199] In some examples of this application, the screening module is specifically used to: when the search target includes pathogen detection, screen out a first result set with the component level being a complete genome; when the first result set meets the third requirement, determine a feedback result based on the first result set; when the first result set does not meet the third requirement, screen out a second result set with the attribute of a reference genome from the first result set, and determine a feedback result based on the second result set.

[0200] In some examples of this application, the target gene sequence is used as a target region to be amplified in a polymerase chain reaction.

[0201] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 7 The device 700 shown can perform Figure 1 The corresponding method embodiments, and the foregoing and other operations and / or functions of each module in device 700 are respectively implemented to achieve Figure 1 For the sake of brevity, the corresponding processes in each method are not described in detail here.

[0202] The apparatus 700 of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the method embodiments in this application can be completed by integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the method disclosed in this application embodiment can be directly embodied as being executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps in the above method embodiments.

[0203] This application also proposes an electronic device. (Reference) Figure 8 The electronic device 600 can be, but is not limited to, the device that performs the above-described method. For example... Figure 6 As shown, the electronic device 600 may include:

[0204] The system includes a memory 610 and a processor 620. The memory 610 stores a computer program 630 and transfers the computer program 630 to the processor 620. In other words, the processor 620 can retrieve and run the computer program 630 from the memory 610 to implement the methods described in the embodiments of this application.

[0205] For example, the processor 620 can be used to execute the steps in the above method according to the instructions in the computer program 630.

[0206] In some embodiments of this application, the processor 620 may include, but is not limited to:

[0207] General-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0208] In some embodiments of this application, the memory 610 includes, but is not limited to:

[0209] Volatile memory and / or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0210] In some embodiments of this application, the computer program 630 may be divided into one or more modules, which are stored in the memory 610 and executed by the processor 620 to perform the method provided in this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 630 in the electronic device.

[0211] like Figure 6 As shown, the electronic device 600 may further include:

[0212] Transceiver 640, which can be connected to processor 620 or memory 610.

[0213] The processor 620 can control the transceiver 640 to communicate with other devices; specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 640 may include a transmitter and a receiver. The transceiver 640 may further include antennas, and the number of antennas may be one or more.

[0214] It should be understood that the various components in the electronic device 600 are connected through a bus system, which includes a data bus, a power bus, a control bus, and a status signal bus.

[0215] According to one aspect of this application, a computer-readable storage medium is provided that stores computer instructions or programs thereon, which, when executed by a computer, enable the computer to perform the methods of the above-described method embodiments. Alternatively, embodiments of this application also provide a computer program product containing instructions that, when executed by a computer, cause the computer to perform the methods of the above-described method embodiments.

[0216] According to another aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method described in the above-described method embodiments.

[0217] In other words, when implemented using software, it can be implemented wholly or partially in the form of a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0218] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0219] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0220] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs. For example, the functional modules in the various embodiments of this application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0221] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0222] The present application's solution will be explained below with reference to embodiments. Those skilled in the art will understand that the following embodiments are for illustrative purposes only and should not be construed as limiting the scope of the present application.

[0223] Example 1: Intelligent Retrieval Application of Cancer-Related Gene Sequences

[0224] The user inputs the query: "What are the reference genes associated with adrenal pheochromocytoma, especially the malignant subtype?" The system identifies the keywords "adrenal pheochromocytoma" and "malignant subtype" through the user interaction module, activates the local Cosmic database, and prompts the user to further confirm the "location" and "subtype" on the front end. After user confirmation, the intelligent agent retrieves cancer-related target reference genes. After the user selects one, it further retrieves cancer-related SNP data. Finally, after the user confirms the candidate gene sequences and SNPs provided by the agent, subsequent primer design experiments are triggered. For the same query question in a cancer scenario, we tested the performance difference between GPT4o and the retrieval target of this invention. We mainly measured the accuracy of the retrieved gene information. GPT4o achieved an average accuracy of 50% in the test data, while this invention achieved an average accuracy of 100%, indicating that this invention is superior. Furthermore, this invention provides sequences, while GPT4o only provides intelligent queries.

[0225] The specific evaluation results are shown in Table 2.

[0226] Table 2

[0227]

[0229] Example 2: Intelligent Retrieval Application of Gene Sequences Related to Hereditary Diseases

[0230] The user inputs the query: "Find gene variant information related to Finnish congenital nephrotic syndrome." The intelligent agent, through multiple rounds of dialogue, clarifies the user's needs and identifies genes related to Finnish congenital nephrotic syndrome and the required variant information. The intelligent agent's search planning module formulates a search strategy and calls databases such as OMIM and ClinVar for data retrieval. The data retrieval module obtains relevant gene and SNP information. When the initial search results are insufficient, the Self-ReAct inference module can proactively optimize the search path, performing a comprehensive search across multiple databases to comprehensively improve data completeness. Finally, detailed gene variant information is presented, including gene function, variant type, and its clinical relevance, which users can use for in-depth research or clinical analysis, supporting targeted detection of genetic diseases. For the same query question in a genetic disease scenario, we tested the performance difference between GPT4o and the search target of this invention. We mainly measured the accuracy of the retrieved gene information. GPT4o achieved an average accuracy of 50% on the test data, while this invention achieved an average accuracy of 100%, indicating that this invention is superior. Furthermore, this invention provides sequences, while GPT4o provides only intelligent queries.

[0231] The specific evaluation results are shown in Table 3.

[0232] Table 3

[0233]

[0234] Example 3: High-throughput screening application of enzyme engineered mutants

[0235] The user inputs the query: "Perform a high-throughput screening process for enzyme-engineered mutants and analyze their related genes." The system identifies the keywords "enzyme-engineered mutants" and "high-throughput screening" through the user interaction module, then guides the user to provide the name of the specific protease to be screened. After the user provides the protein name, the intelligent agent's search planning module formulates a search strategy, calling the UniProt and COSMIC databases to obtain the sequence information of relevant proteins. The data retrieval module obtains relevant information for multiple proteins, such as DNA sequence information, amino acid sequence information, and protein function. When the search results do not meet the requirements, the intelligent agent's Self-ReAct inference module can proactively adjust the search parameters and path, re-search and integrate the results from multiple protein databases. The intelligent agent's result verification module presents the DNA sequence information, amino acid sequence information, and protein function in the form of charts and tables. After the user confirms and selects the corresponding protein sequence information, it can be used to generate the multiplex PCR primer pool required for targeted sequencing and perform targeted sequencing for high-throughput screening experiments of enzyme mutants. For the same query problem in the enzyme engineering (protein) scenario, we tested the performance difference between GPT4o and the search targets of this invention. We evaluated the accuracy of GPT4o and this invention based on the database sequence access number (NC number), the websites that can provide sequences, and the sequences that can be retrieved. GPT4o achieved an average accuracy of 40% in the test data, while this invention achieved an average accuracy of 100%, indicating that this invention is superior. Furthermore, this invention can provide sequences, while GPT4o provides intelligent queries.

[0236] The specific evaluation results are shown in Table 4.

[0237] Table 4

[0238]

[0239] Example 4: Application of Pathogen Genome Detection

[0240] The user inputs the query: "Design a multiplex PCR primer pool for whole-genome sequencing of the novel coronavirus." The intelligent agent identifies the keywords "novel coronavirus" and "whole genome" through the user interaction module and guides the user to provide specific virus names or strain information. After the user provides relevant information, the search planning module formulates a search strategy and calls the NCBI Genome database to obtain relevant genome information. The data retrieval module obtains the viral genome sequence, and the Self-ReAct inference module adjusts the query parameters and re-retrieves and integrates results from multiple databases if the initial search results are incomplete. The result verification module displays the search results in a detailed report, including annotation information related to the viral genome, such as assembly level, isolate, source, species name, strain name, and database access number. After user confirmation, the results can be used for pathogen detection and PCR experimental analysis. For the same query problem in the pathogen detection scenario, we tested the performance difference between GPT4o and the search target of this invention. We evaluated the accuracy of GPT4o and this invention based on the database sequence access number (NC number), the websites that can provide sequences, and the sequences that can be retrieved. GPT4o achieved an average accuracy of 40% in the test data, while the present invention achieved an average accuracy of 100%, indicating that the present invention is superior. In addition, the present invention can provide sequences, while GPT4o provides intelligent queries.

[0241] The specific evaluation results are shown in Table 5.

[0242] Table 5

[0243]

[0244]

[0245] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0246] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application without departing from the principles and spirit of this application.

Claims

1. A method for retrieving gene information, characterized in that, The method includes: Search suggestions are received through a large language model, and the search suggestions include suggestions related to the target gene. The search target of the search suggestion words is determined by the large language model, wherein different search targets have a first mapping relationship with different scene agents, and each scene agent encapsulates the processing logic corresponding to the search target, the processing logic including the main database to be searched; A retrieval strategy is generated by a target scene agent that has a first mapping relationship with the retrieval target, wherein the retrieval strategy includes retrieval keywords and a target master database, and the target master database is the master database in the logic encapsulated by the target scene agent; The target master database is searched based on the search keywords to determine the search results for the target gene.

2. The method according to claim 1, characterized in that, There is a second mapping relationship between different search targets and different prompt word templates; after determining the search target of the search prompt words through the large language model, the method further includes: Obtain a target suggestion word template that has a second mapping relationship with the search target, wherein the target suggestion word template includes at least one pair of example information, wherein each pair of example information includes a prior search suggestion word with the search target and its corresponding prior search keyword; Based on the target suggestion term template and the current search suggestion term, the search keywords corresponding to the current search suggestion term are determined.

3. The method according to claim 1, characterized in that, The step of generating a retrieval strategy through a target scene intelligent agent that has a first mapping relationship with the retrieval target includes: The search parameters are determined based on at least one of the following information, wherein the search parameters include search filtering conditions, and the information includes: default parameters associated with the target search scenario and received user input parameters; A search strategy is determined based on the search keywords, the target master database, and the search parameters.

4. The method according to claim 1, characterized in that, If it is determined that the search result does not meet the first requirement, the method further includes: The large language model is used to identify areas for improvement in the search results. Based on the search strategy, the search results, and the areas for improvement, an optimized search strategy is determined.

5. The method according to claim 4, characterized in that, The areas for improvement include at least one of the following: adjusting search parameters, determining auxiliary databases to expand the search scope, optimizing the query route when the search scope includes at least two databases, and optimizing the search keywords; The query route includes the retrieval order of different databases.

6. The method according to any one of claims 1-5, characterized in that, The target master database is searched based on the search keywords to determine the search results for the target gene sequence, including: Determine the vectorized database corresponding to the target master database, wherein the vectorized database includes multiple document vectors, and the document vectors are obtained by vectorizing the documents in the target master database respectively; Determine the retrieval vector corresponding to the search keywords; Calculate the vector similarity between the retrieval vector and the document vectors in the vectorized database; Based on the vector similarity, the search results for the target gene sequence are determined, wherein the search results are determined based on document vectors whose vector similarity is greater than a preset threshold, or the search results are determined based on the top preset number of document vectors with the highest vector similarity. or, Calculate the matching degree between the search keywords and the documents in the target master database; Based on the matching degree, search results for the target gene sequence are determined, wherein the search results are determined based on documents with a matching degree greater than a preset threshold, or the search results are determined based on the top preset number of documents with the highest matching degree.

7. The method according to any one of claims 1-5, characterized in that, If the retrieval process is based on non-vector search keywords, and the search results do not meet the second requirement, the method further includes at least one of the following: Determine the retrieval vector corresponding to the retrieval keyword, and perform a retrieval in a vectorized database based on the retrieval vector, wherein the vectorized database is obtained by vectorizing the documents in the database to be retrieved; Identify synonyms of the search keywords, use the synonyms as search keywords, and perform a search based on the search keywords.

8. The method according to any one of claims 1-5, characterized in that, There is a third mapping relationship between different search targets and different filtering mechanisms; If the search results meet the third requirement, the method further includes: A target filtering mechanism that obtains a third mapping relationship with the search target; The search results are filtered based on the target filtering mechanism, and the filtered search results are used to provide feedback to the user.

9. A gene information retrieval device, characterized in that, include: The receiving module is used to receive search suggestions through a large language model, wherein the search suggestions include suggestions related to the target gene; The first determining module is used to determine the retrieval target of the retrieval prompt words through the large language model, wherein different retrieval targets have a first mapping relationship with different scene intelligent agents, and each scene intelligent agent encapsulates the processing logic corresponding to the retrieval target, the processing logic including the main database to be retrieved; The generation module is used to generate a retrieval strategy through a target scene agent that has a first mapping relationship with the retrieval target, wherein the retrieval strategy includes retrieval keywords and a target master database, and the target master database is the master database in the logic encapsulated by the target scene agent; The retrieval module is used to search the target master database based on the search keywords and determine the retrieval results for the target gene sequence.

10. An electronic device, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is configured to execute the computer program to implement the method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The storage medium includes computer instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1-8.

12. A computer program product, characterized in that, The computer program product includes computer instructions that, when some or all of the computer instructions are run on a computer, cause the method as described in any one of claims 1-8 to be performed.