Large-model-driven intelligent biological research method and system

Abstract

The invention discloses a large-model-driven intelligent biological research method and system, and aims to realize continuous optimization of a model and improve the efficiency and reliability of biological science research through a feedback mechanism verified by an experiment. According to the method, a closed-loop process is constructed and comprises the steps that firstly, data processing is optimized through a large model, and high quality and diversity of data are ensured; then, the large model participates in an analysis and prediction task of bioinformatics, and a high-precision prediction result is provided through an advanced algorithm and reasoning ability; based on the prediction result, large model optimization experiment design covers experiment priority ranking, experiment parameter optimization and automatic experiment assistance; then experimental verification is carried out; and finally, an experimental result is deeply analyzed through a large model, and if the experimental result is inconsistent with a prediction result, the experimental result is converted into an input format of the model for further training and optimization of the model, so that the accuracy and adaptability of model prediction are continuously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of biological information, in particular to a large model driven intelligent biological research method and system. BACKGROUND

[0002] In the field of life science research, data acquisition and experimental validation often require expensive manpower, computing resources and sophisticated instruments. Traditional biological research methods usually collect data sets before model training and evaluate the model offline on these data sets. However, this method has the following problems, first, the scale of the training data set is limited, which is difficult to meet the training needs of large-scale parameter models; second, there is a lack of real-world feedback, which makes the evaluation of model capability unreliable; third, the experimental design and model prediction are seriously disconnected, which cannot effectively utilize model prediction to guide experiments, nor can it utilize experimental results to feedback model training.

[0003] In recent years, large language models have achieved great success, demonstrating their strong ability in text understanding and generation. The achievements of these models have inspired the potential of applying large models in the field of life sciences. The present application aims to explore the application of large models in life science research. By constructing a closed-loop system of model training and experimental validation, the application ability of models in life science research is enhanced, and scientific discovery is promoted. Through this closed-loop system, researchers can achieve data-driven experimental design, real-time feedback model optimization and automated experimental execution, solving the problems of data limitation and insufficient feedback in traditional methods, and ultimately accelerating the process of scientific research. SUMMARY

[0004] In view of the problem that current model prediction and real-world experimental validation cannot form an effective closed loop, the present application proposes a large model driven intelligent biological research method and system.

[0005] The purpose of the present application is achieved by the following technical solutions: The present application provides a large model driven intelligent biological research method, comprising the following steps: Step 1: extracting biological data using a large model, and pre-processing, data enhancement and feature extraction of the biological data; Step 2: based on the biological data processed in step 1, the large model participates in the analysis and prediction task of bioinformatics, and obtains the prediction result; Step 3: based on the prediction result, using the large model to optimize the experimental design, covering experimental priority sorting, experimental parameter optimization and automated experimental assistance; Step 4: performing experiments according to the optimized experimental design; Step 5: In-depth analysis of experimental results by large model; if the experimental results are inconsistent with the predicted results, convert the experimental results into the input format of the model for further training and optimization of the model; if the experimental results are consistent with the predicted results, extract the mechanism behind the experimental results from the literature data by the large model, and output the deep understanding of the experimental results.

[0006] Further, step 1 includes: Collect biological data from various sources, including public databases, experimental results, and literature, and convert experimental data and database data into a format suitable for the model; use the large model to extract information related to biological data from scientific literature to build a knowledge graph or supplement the data set; simulate experimental data or supplement small sample data sets using generative capabilities; use the large model to detect and correct missing values, outliers, and noise in the data; use the deep learning capabilities of the large model to automatically extract important features from the data and reduce redundant information.

[0007] Further, the format suitable for the model includes sequence format, structure format, and graph data format; the information related to biological data includes publicly available protein sequences, functional attributes, and molecular characteristic information; the important features in the data include structural features, physical and chemical features, dynamic features, and evolutionary features.

[0008] Further, step 2 includes: The large model selects appropriate models for training and prediction according to task requirements, provides prior knowledge or rules to guide the training of the model, and uses the features extracted from biological data by the large model to help other models understand biological data; the large model participates in the prediction task and directly predicts based on biological data or knowledge graph by designing appropriate prompts.

[0009] Further, step 3 includes: According to the model prediction results and the availability of experimental resources, prioritize experiments to ensure that critical experiments are performed first; automatically optimize experimental parameters through model and historical data analysis; design automated experimental processes to perform repetitive experimental steps using robots and automated equipment; provide suggestions on experimental methods and techniques by human experts.

[0010] Further, step 4 includes: Perform experiments according to the optimized experimental design to ensure strict control of experimental conditions; use high-precision equipment to collect experimental data to ensure data integrity and accuracy.

[0011] Further, step 5 includes: The large model is used to deeply analyze the experimental results, and the consistency with the predicted results is verified; if the experimental results are inconsistent with the predicted results, the result data is converted into the input format of the model, and the model is retrained and optimized, and the parameters and structure of the model are continuously adjusted through the feedback mechanism.

[0012] The application also provides a large model-driven intelligent biological research system for realizing the above method, comprising: A data preprocessing module is configured to extract biological data using a large model, and to preprocess, enhance data, and extract features of the biological data; An analysis and prediction module is configured to participate in analysis and prediction tasks of bioinformatics based on the processed biological data, and to obtain prediction results using a large model; An experiment design module is configured to optimize experiment design based on the prediction results using a large model, including experiment priority sorting, experiment parameter optimization, and automated experiment assistance; An experiment execution module is configured to execute experiments according to the optimized experiment design; A result analysis and feedback module is configured to deeply analyze experimental results using a large model; if the experimental results are inconsistent with the predicted results, the experimental results are converted into the input format of the model for further training and optimization of the model; if the experimental results are consistent with the predicted results, the mechanism behind the experimental results is extracted from literature data using a large model, and a deep understanding of the experimental results is output.

[0013] The application also provides an electronic device comprising a memory and a processor, wherein the memory is coupled to the processor; the memory is configured to store program data, and the processor is configured to execute the program data to realize the above-mentioned large model-driven intelligent biological research method.

[0014] The application also provides a computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to realize the above-mentioned large model-driven intelligent biological research method.

[0015] The application has the following advantages: the application provides a large model-driven intelligent biological research method, realizes an effective closed loop between model training and real-world experiment verification, improves the training and evaluation efficiency of the model, and enhances the application ability of the model in scientific research; realizes data-driven experiment design, real-time feedback model optimization, and automated experiment execution, solves the data limitation and feedback deficiency problem in traditional methods. The intelligent method not only improves the efficiency of biological research and reduces the cost, but also promotes the progress of scientific discovery through continuous feedback and optimization mechanism. Its significant scientific and commercial value lies in accelerating the research cycle, reducing experimental resource waste, and providing more reliable research results. The application provides an innovative solution for biological experimental research, and has wide application prospect and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the large model-driven intelligent biological research method proposed in the present invention; Figure 2 This is a schematic diagram of the structure of the large model-driven intelligent biological research system provided by the present invention; Figure 3 It is a schematic diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0017] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of systems and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0018] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a," "the," and "the" used in this invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0019] It should be understood that although the terms "first," "second," "third," etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0020] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in combination with specific implementation cases and with reference to the accompanying drawings.

[0021] See also Figure 1 The present invention proposes a large-scale model-driven intelligent biological research method, the main steps of which include: Step 1: Use large models to optimize data processing and ensure high data quality and diversity.

[0022] Specifically, biological data is collected from various sources, including public databases, experimental results, and literature, and is converted into formats suitable for models (such as sequence format, structure format, and graph data format). Large models are used to extract information related to biological data from scientific literature, such as published protein sequences, functional attributes, and molecular feature information. When integrating multi-source heterogeneous data (such as combining structure, sequence, and literature conclusions) or analyzing cross-study correlations, knowledge graphs are constructed. When there is insufficient original data, data sets are supplemented. By generating experimental data or supplementing small sample data sets, the problem of insufficient data is alleviated. Large models are used to detect and correct missing values, outliers, and noise in data. The deep learning capabilities of large models are used to automatically extract important features from data (such as structural features, physical and chemical features, dynamic features, and evolutionary features), reducing redundant information. Structural features can include CDR loop geometric parameters (length, angle), disulfide bond positions, and solvent accessible surface area (dSASA) of antigen binding interface residues. Physical and chemical features can include hydrophobicity and charge distribution (which affects stability). Dynamic features can include RMSF values (flexible regions) from molecular dynamics (MD) simulations. Evolutionary features can include conserved sites in multiple sequence alignments (MSA).

[0023] For example, in the context of nanobody design, public databases include structural databases such as PDB (Protein Data Bank) and SAbDab (Structural Antibody Database), which contain 3D structures, sequences, and antigen binding information for nanobodies, as well as sequence databases such as UniProt, which provide amino acid sequences and genetic data for nanobodies. Experimental results include data obtained from high-throughput sequencing, X-ray crystallography / cryo-EM, surface plasmon resonance, and mutation scanning. Literature includes research papers, patent application texts, and other data. Information related to biological data includes published nanobody sequences, uses, mutation sites, key residues, binding affinities, and other information.

[0024] The data formats available for the model include: sequence format, such as extracting protein sequences from PDB files through tools like BioPython's PDBParser, and the large model can assist in writing sequence format conversion code; structure format, such as optimizing structures with Rosetta or OpenMM, and the large model can help generate structure processing scripts; graph data format, representing nanobody-antigen complexes as graphs, tools like PyG or DGL, and the large model can assist in designing graph construction strategies and implementing code. In these data format conversion processes, the large model can provide the following support: automatically generating format conversion code, optimizing data processing workflows, providing error troubleshooting suggestions, recommending appropriate processing tools and parameters, and explaining the characteristics and application scenarios of different data formats.

[0025] The types of missing data in the data include sequence missing, structure missing, and experimental data missing (such as binding affinity), which can be filled by generating models or repaired based on multi-source data integration, such as predicting missing CDR region sequences through protein language models; detecting outliers through energy distribution or RMSD threshold, such as analyzing the distribution of the training set through the large model, dynamically adjusting the RMSD threshold, and correcting through molecular dynamics or experimental determination; noise includes experimental noise and computational noise, which can be handled using tools like HDOCK.

[0026] The large model is combined with bioinformatics tools to improve the efficiency and expand the functionality of these tools.

[0027] Step 2: The large model participates in the analysis and prediction tasks of bioinformatics, providing high-precision prediction results through advanced algorithms and reasoning capabilities.

[0028] Specifically, the large model selects appropriate models for training and prediction based on task requirements, such as selecting DiffAb or ADesigner models for CDR sequence generation and AlphaFold2 or IgFold models for structure prediction; the large model provides prior knowledge or rules to guide the training of models and helps other models understand biological data by extracting features from biological data; the large model can directly participate in prediction tasks by designing appropriate prompts and directly predicting based on biological data or knowledge graphs.

[0029] For example, in nanobody design, the rules provided by the large model can include the following aspects: structural constraints (such as length restrictions on CDRH3), physical and chemical rules (such as hydrophobicity and charge complementarity requirements for binding interface residues), energy thresholds, and evolutionary conservation considerations.

[0030] For example, the prompts are as follows: For affinity prediction, the prompt design is: "Predict ΔG (kcal / mol) for the nanobody-antigen complex in PDB:8PMH, considering mutations H52Y and S73R". The corresponding Chinese meaning of this prompt word is: "Please predict the change of binding free energy (ΔG, unit: kcal / mol) of the nanobody-antigen complex in the PDB database with the number 8PMH, and consider the influence of mutating the 52nd histidine (H) to tyrosine (Y) and the 73rd serine (S) to arginine (R) in the nanobody".

[0031] For the CDRH3 generation task, the prompt design is: "Generate a 15-residue CDRH3 sequence for a VHH binding to KRAS G12D, with a disulfide bond between positions 3 and 12". The corresponding Chinese meaning of this prompt word is: "Please generate a 15-residue CDRH3 sequence for a VHH binding to KRAS G12D, with a disulfide bond between positions 3 and 12".

[0032] Step 3, based on the prediction results, the large model optimizes the experimental design, covering experimental priority ranking, experimental parameter optimization and automated experimental assistance.

[0033] Specifically, according to the model prediction results and the availability of experimental resources, the experiments are prioritized to ensure that key experiments are performed first and unnecessary experiments are reduced; through model and historical data analysis, experimental parameters are automatically optimized to improve experimental efficiency and success rate; design automated experimental processes, use robots and automated equipment to perform repetitive experimental steps to improve experimental efficiency; human experts participate to provide suggestions on experimental methods and techniques to ensure the scientificity and rationality of experimental design.

[0034] By using LLM (Large Language Model) agents, a system capable of designing, planning, optimizing and executing scientific experiments is developed.

[0035] Exemplarily, in the expression condition optimization experiment, according to the soluble expression amount at different induction temperatures in the historical data, the relationship between temperature and expression amount is modeled by Gaussian process regression, and then the recommended optimal temperature expression experiment is performed.

[0036] Step 4, perform experimental verification.

[0037] Specifically, the experiment is performed according to the optimized experimental design, ensuring strict control of the experimental conditions; high-precision equipment is used to collect experimental data, ensuring the integrity and accuracy of the data. High-precision equipment can be selected from X-ray crystal diffraction, cryo-EM, surface plasmon resonance (SPR), small-angle X-ray scattering (SAXS), isothermal titration calorimetry (ITC), microscale thermophoresis (MST), etc.

[0038] Step 5, using a large model to analyze the experimental results in depth, verify the consistency with the predicted results; if the experimental results are inconsistent with the predicted results, the result data is converted into the input format of the model, retrained and optimized, and the parameters and structure of the model are adjusted through the feedback mechanism, so as to continuously improve the accuracy and adaptability of the model prediction.

[0039] Illustratively, when the experiment verifies that the actual binding affinity and structural stability of the nanobody have significant deviations from the model prediction, the system starts the following optimization process: data conversion and labeling: convert the experimental data into a format that the model can process, model retraining: incremental training: use the experimental data as new samples to fine-tune the model or adversarial training: introduce adversarial samples (such as low-affinity design), enhance the model's ability to identify "failed cases". The model parameters and structure can be optimized using active learning, such as using a large model to identify high-uncertainty designs (e.g. samples with large prediction ΔG variance), and preferentially recommending these samples for experimental verification to maximize information gain.

[0040] If the experimental results are consistent with the predicted results, the large model extracts the mechanism behind the experimental results from a large amount of literature data, providing a deep understanding of the experimental results.

[0041] Illustratively, when the prediction is consistent with the experimental results, the large model extracts the relevant mechanism from the literature (such as the CDR3 protruding structure enhancing antigen binding), and generates an explainability report.

[0042] The large model-driven intelligent biological research system provided by the present application is described below, and the large model-driven intelligent biological research system described below can be referred to in conjunction with the large model-driven intelligent biological research method described above.

[0043] Figure 2 The structure diagram of the large model-driven intelligent biological research system provided by the present application is shown in Figure 2 As shown in the figure, the large model-driven intelligent biological research system provided by the present application includes: A data preprocessing module for extracting biological data using a large model and preprocessing, data enhancement and feature extraction of the biological data; The analysis and prediction module is used to participate in the analysis and prediction tasks of bioinformatics based on the processed biological data and obtain the prediction results; Experimental design module, which is used to optimize experimental design based on prediction results using large models, covering experimental priority sorting, experimental parameter optimization and automated experimental assistance; Experiment execution module, used to execute experiments according to the optimized experimental design; The result analysis and feedback module is used to deeply analyze the experimental results through the large model; if the experimental results are inconsistent with the predicted results, the experimental results are converted into the model input format for further training and optimization of the model; if the experimental results are consistent with the predicted results, the large model is used to extract the mechanism behind the experimental results from the literature data and output a deep understanding of the experimental results.

[0044] Corresponding to the embodiment of the aforementioned large model driven intelligent biological research method, the embodiment of the present application further provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned large model driven intelligent biological research method. Figure 3 As shown, it is a hardware structure diagram of any device with data processing capability in the large model driven intelligent biological research method provided in the embodiment of the present application, except Figure 3 In addition to the processor, memory, DMA controller, disk, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0045] Corresponding to the aforementioned embodiment of the large model-driven intelligent biological research method, an embodiment of the present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the large model-driven intelligent biological research method in the above embodiment is implemented.

[0046] The computer readable storage medium can be an internal storage unit of any of the aforementioned data processing capable devices, such as a hard disk or a memory. The computer readable storage medium can also be any of the aforementioned data processing capable devices, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. Further, the computer readable storage medium can also include both an internal storage unit of any of the aforementioned data processing capable devices and an external storage device. The computer readable storage medium is used to store the computer program and other programs and data required by the aforementioned data processing capable devices, and can also be used to temporarily store data that has been output or will be output.

[0047] The above merely provides the preferred embodiments of the present application, but not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0048] The above embodiments are only used to illustrate the design ideas and features of the present application, and the purpose is to enable those skilled in the art to understand the present application and implement it, and the protection scope of the present application is not limited to the above embodiments. Therefore, any equivalent change or modification made according to the disclosed principles and design ideas of the present application shall be included in the protection scope of the present application.

Claims

1. A large-scale model-driven intelligent biological research method, characterized in that: The following steps are involved: Step 1: Use the large model to extract biological data and perform preprocessing, data enhancement and feature extraction on the biological data; Step 2: Based on the biological data processed in step 1, the large model participates in the analysis and prediction task of bioinformatics to obtain the prediction results; Step 3: Based on the prediction results, use the big model to optimize the experimental design, including experimental priority sorting, experimental parameter optimization and automated experimental assistance; Step 4: Execute the experiment according to the optimized experimental design; Step 5: Use the big model to deeply analyze the experimental results. If the experimental results are inconsistent with the predicted results, convert the experimental results into the model input format for further training and optimization. If the experimental results are consistent with the predicted results, use the big model to extract the mechanism behind the experimental results from the literature data and output a deep understanding of the experimental results.

2. The large model-driven intelligent biological research method according to claim 1, characterized in that: Step 1 includes: Collect biological data from a variety of sources, including public databases, experimental results, and literature, and convert experimental data and database data into a format suitable for the model; use large models to extract information related to biological data from scientific literature to build knowledge graphs or supplement data sets; simulate experimental data or supplement small sample data sets through generative capabilities; use large models to detect and correct missing values, outliers, and noise in the data; use the deep learning capabilities of large models to automatically extract important features in the data and reduce redundant information.

3. The large model-driven intelligent biological research method according to claim 2, characterized in that: The formats applicable to the model include sequence format, structure format, and graph data format; the information related to the biological data includes published protein sequences, functional properties and molecular characteristic information; the important features in the data include structural features, physical and chemical features, dynamic features, and evolutionary features.

4. The large model-driven intelligent biological research method according to claim 1, characterized in that: Step 2 includes: The large model selects a suitable model for training and prediction according to task requirements. The large model provides prior knowledge or rules to guide model training, and uses the features extracted from biological data by the large model to help other models understand biological data. The large model participates in prediction tasks, and by designing appropriate prompts, the large model makes predictions directly based on biological data or knowledge graphs.

5. The large model-driven intelligent biological research method according to claim 1, characterized in that: Step 3 includes: Prioritize experiments based on model prediction results and the availability of experimental resources to ensure that key experiments are carried out first; automatically optimize experimental parameters through model and historical data analysis; design automated experimental processes and use robots and automated equipment to perform highly repeatable experimental steps; and provide advice on experimental methods and techniques through human experts.

6. The large model-driven intelligent biological research method according to claim 1, characterized in that: Step 4 includes: Perform experiments according to optimized experimental designs to ensure strict control of experimental conditions; use high-precision equipment to collect experimental data to ensure data integrity and accuracy.

7. The large model-driven intelligent biological research method according to claim 1, characterized in that: Step 5 includes: Use a large model to conduct in-depth analysis of the experimental results to verify their consistency with the predicted results; if the experimental results are inconsistent with the predicted results, convert the result data into the input format of the model, retrain and optimize, and continuously adjust the parameters and structure of the model through the feedback mechanism.

8. A large model driven intelligent biological research system implementing the method of claim 1, characterized in that: include: Data preprocessing module, used to extract biological data using large models, and perform preprocessing, data enhancement and feature extraction on the biological data; The analysis and prediction module is used to participate in the analysis and prediction tasks of bioinformatics based on the processed biological data and obtain the prediction results; Experimental design module, which is used to optimize experimental design based on prediction results using large models, covering experimental priority sorting, experimental parameter optimization and automated experimental assistance; Experiment execution module, used to execute experiments according to the optimized experimental design; The result analysis and feedback module is used to deeply analyze the experimental results through the large model; if the experimental results are inconsistent with the predicted results, the experimental results are converted into the model input format for further training and optimization of the model; if the experimental results are consistent with the predicted results, the large model is used to extract the mechanism behind the experimental results from the literature data and output a deep understanding of the experimental results.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the large model-driven intelligent biological research method described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the large model-driven intelligent biological research method as described in any one of claims 1 to 7 is implemented.

Images