An agent-based automated protein property prediction method and system

Abstract

The present application relates to the field of information technology, and in particular to an automatic protein property prediction method and system based on an agent. The method and system comprise: receiving a user input protein analysis task instruction, decomposing the analysis task instruction; task analysis is carried out on the decomposed analysis task instruction, a custom function for protein understanding task is accessed, the core requirement of the protein analysis task instruction is identified, and the required analysis task type is determined; according to the analysis task type, the custom function is accessed, the analysis and prediction of protein data are completed based on a general language model, and reasoning results are generated. The present application can automatically execute protein analysis tasks, realize automatic processing from user instructions to analysis results, thereby significantly improving protein analysis efficiency and reducing user usage threshold.

Description

Technical Field

[0001] This invention relates to the field of information technology, and more specifically, to an automated method and system for predicting protein properties based on intelligent agents. Background Technology

[0002] In computational biology, protein sequences are considered data sequences akin to "biological language." Researchers have employed techniques such as masked language modeling to develop various protein language models (e.g., ESM, ProtTrans, ProGPT2). These models are trained on large amounts of protein data to extract semantic information from protein sequences. However, current protein language models struggle to meet the demands of task automation and lack human-computer interaction capabilities during protein analysis tasks.

[0003] A novel multi-agent framework, ProtAgents, has been introduced to collaboratively handle complex tasks such as protein structure analysis and physical simulations. This framework can support protein design and analysis to some extent. However, ProtAgents and similar implementations still require a high level of user expertise and cannot fully realize the natural language command and automation of protein tasks. Other existing solutions, such as PMC-LLaMA and DrugAssist, while performing well in biomolecular analysis and task planning, primarily focus on single tasks and are ill-suited to the needs of multi-task collaborative analysis.

[0004] In summary, the shortcomings of existing technologies are as follows:

[0005] 1. Insufficient task automation: Current protein language models require extensive preprocessing and tuning, and users must manually perform data preparation and model configuration processes, resulting in low task execution efficiency.

[0006] 2. Limited instruction comprehension: Existing protein models have shortcomings in instruction comprehension and natural language interaction, making it difficult to automatically execute complex tasks based on user instructions.

[0007] 3. Lack of multi-task collaboration: Existing multi-agent frameworks have limited capabilities in handling multi-task collaboration for protein tasks, and cannot efficiently handle various tasks such as protein property prediction, protein-drug interaction, and protein-protein interaction. Summary of the Invention

[0008] This invention provides an automated protein property prediction method and system based on intelligent agents, which at least addresses the technical problem of insufficient automation in existing protein analysis tasks.

[0009] According to an embodiment of the present invention, an automated protein property prediction method based on an intelligent agent is provided, comprising the following steps:

[0010] S101: Receives the protein analysis task instruction input by the user and decomposes the analysis task instruction;

[0011] S102: Perform task parsing on the decomposed analysis task instructions, access the custom functions for protein understanding tasks, identify the core requirements of the protein analysis task instructions, and determine the required analysis task type.

[0012] S103: According to the analysis task type, access the custom function, complete the parsing and prediction of protein data based on the general language model, and generate inference results.

[0013] Furthermore, the method also includes:

[0014] Analyze the generated inference results, access custom functions, and calculate task metrics based on the inference results.

[0015] Furthermore, the method also includes:

[0016] Access customizable features and perform visual operations based on task metrics.

[0017] Furthermore, in step S102, the analysis task types include: protein property prediction, protein-drug interaction, and protein-protein interaction.

[0018] Furthermore, in step S102, the protein analysis task is decomposed into multiple sub-tasks and executed in a coordinated manner. The complex task requirements are parsed by utilizing the natural language understanding capabilities of GPT-4, ensuring the accuracy of task decomposition and execution.

[0019] Furthermore, the general language model includes a general large language model and / or a general protein language model.

[0020] Furthermore, visualization includes generating visualization charts, such as ROC curves and scatter plots.

[0021] According to another embodiment of the present invention, an automated protein property prediction system based on an intelligent agent is provided, comprising:

[0022] The user agent is used to receive protein analysis task instructions input by the user and decompose the analysis task instructions.

[0023] The chat management agent is used to parse the decomposed analysis task instructions, access custom functions for protein understanding tasks, identify the core requirements of protein analysis task instructions, and determine the required analysis task type.

[0024] The reasoning agent is used to access custom functions according to the type of analysis task, to parse and predict protein data based on a general language model, and to generate reasoning results.

[0025] Furthermore, the system also includes:

[0026] Evaluate the agent to analyze the generated inference results, access custom functions, and calculate task metrics based on the inference results.

[0027] Furthermore, the system also includes:

[0028] Visual agents are used to access custom functions and perform visual operations based on task metrics.

[0029] Furthermore, the inference agent includes multiple customized analysis function modules, which automatically call appropriate data processing methods to generate prediction results for different protein analysis tasks.

[0030] A storage medium storing program files capable of implementing any of the above-mentioned agent-based automated protein property prediction methods.

[0031] A processor for running a program, wherein the program executes any of the above-mentioned agent-based automated protein property prediction methods during runtime.

[0032] The automated protein property prediction method and system based on intelligent agents in this invention can automatically execute protein analysis tasks and realize automated processing from user instructions to analysis results, thereby significantly improving protein analysis efficiency and reducing the user threshold. Attached Figure Description

[0033] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0034] Figure 1 This is a flowchart of the automated protein property prediction method based on intelligent agents according to the present invention;

[0035] Figure 2 This is a diagram of the ProtChat system architecture of the present invention;

[0036] Figure 3 This is a schematic diagram of the evaluation and visualization process in this invention;

[0037] Figure 4 This is a preferred module diagram of the intelligent agent-based automated protein property prediction system of the present invention. Detailed Implementation

[0038] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0039] 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 a non-exclusive inclusion; for example, a process, method, system, product, or apparatus 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 apparatus.

[0040] Example 1

[0041] With the development of large language models (LLMs), protein language models (PLLMs) have shown great potential in bioinformatics for processing protein data. However, existing PLLMs cannot effectively handle complex protein analysis tasks, especially when users without computational background need to perform protein analysis via natural language commands. Protein language models have significant shortcomings in task automation and the understanding and execution of task commands. Existing protein analysis workflows are complex, often requiring manual preprocessing, script building, and model tuning, leading to inefficiency. Therefore, there is an urgent need for an automated system that integrates natural language understanding and protein analysis functions to improve the efficiency of protein attribute prediction, protein-drug interaction, and protein-protein interaction tasks.

[0042] This invention aims to develop an automated protein property prediction method based on intelligent agents—ProtChat. By integrating the task planning capabilities of the large-scale language model GPT-4 with the analytical capabilities of a protein language model, it completes the parsing and prediction of protein data based on a general large-scale language model and a general language model. Combining the advantages of the large-scale language model ChatGPT and the general language model, it provides life science researchers outside of computation with a framework for accurate protein physicochemical property prediction calculations using simple natural language. This invention can automatically execute protein analysis tasks, achieving automated processing from user commands to analysis results. ProtChat supports protein property prediction, protein-drug interaction, and protein-protein interaction analysis, and automatically generates visualization results, thereby significantly improving protein analysis efficiency and lowering the user threshold.

[0043] According to an embodiment of the present invention, an automated protein property prediction method based on an intelligent agent is provided, see [link to relevant documentation]. Figure 1 This includes the following steps:

[0044] S101: Receives the protein analysis task instruction input by the user and decomposes the analysis task instruction;

[0045] S102: Perform task parsing on the decomposed analysis task instructions, access the custom functions for protein understanding tasks, identify the core requirements of the protein analysis task instructions, and determine the required analysis task type.

[0046] S103: According to the analysis task type, access the custom function, complete the parsing and prediction of protein data based on the general language model, and generate inference results.

[0047] The agent-based automated protein property prediction method in this invention can automatically execute protein analysis tasks, realize automated processing from user instructions to analysis results, thereby significantly improving protein analysis efficiency and reducing the user threshold.

[0048] The core technical solution of this invention includes the following modules:

[0049] User Proxy module: This module is responsible for receiving user input commands and passing them to the Chat Manager to generate task plans.

[0050] Inference Agent: Responsible for performing protein analysis tasks, calling the MASSA model to process protein data, and supporting various tasks such as protein attribute prediction, protein-drug interaction analysis, and protein-protein interaction analysis.

[0051] Evaluation Agent: Analyzes and evaluates the inference results, calculates and analyzes task metrics (such as accuracy, AUC, etc.), and feeds the results back to the "Chat Manager".

[0052] Visualization Agent: Visualizes the analysis results of a task and generates user-friendly results displays such as images.

[0053] ProtChat's multi-agent system can dynamically allocate and coordinate tasks, enabling command-driven automated protein analysis. Through simple user command interaction, the system can automatically identify the required tools and task flows, thereby greatly simplifying the complexity and reducing the barrier to entry for protein analysis.

[0054] The technical solution of the present invention is described in detail below:

[0055] The ProtChat system of this invention automates protein analysis tasks through multiple collaborative intelligent agent modules. See also... Figure 2 This diagram showcases the overall architecture of the ProtChat system, including the relationships between the user agent, chat manager, inference agent, evaluation agent, and visualization agent. The diagram clearly illustrates the functional division of each module and their collaborative methods. An example of a user interface inputting tasks via natural language commands, along with the visual results provided to the user after task completion, demonstrates ProtChat's user-friendly interface.

[0056]

[0057] Table 1

[0058]

[0059]

[0060] Table 2

[0061] Based on Table 1-2, the following are the detailed technical solutions for each agent module in the ProtChat system:

[0062] User Proxy module:

[0063] This module serves as the interface between the user and the system, receiving protein analysis instructions from the user and supporting instruction input in natural language. The user agent passes the instructions to the "Chat Manager" module for task parsing, identifies the core requirements of the instructions, and determines the required analysis task type (such as protein property prediction, protein-drug interaction, protein-protein interaction, etc.).

[0064] Chat Manager:

[0065] The chat manager receives instructions from the user agent and assigns tasks to the appropriate agents. Its core function is to break down protein analysis tasks into multiple sub-tasks and coordinate the sequential execution of these sub-tasks by the agents. The chat manager leverages GPT-4's natural language understanding capabilities to parse complex task requirements, thereby ensuring the accuracy of task decomposition and execution.

[0066] Inference Agent:

[0067] The inference agent is responsible for performing the analysis of protein data, parsing and predicting protein data based on a general language model. The inference agent includes several customized analysis function modules, such as "Analyze_Protein_Property", "Analyze_Protein_Drug_Interaction", and "Analyze_Protein_Protein_Interaction". These modules can automatically call appropriate data processing methods for different protein analysis tasks, generate prediction results, and return them to the chat manager.

[0068] Evaluation Agent:

[0069] The evaluation agent is responsible for analyzing the predictions generated by the inference agent and calculating task metrics (such as accuracy, ROC curve, RMSE, etc.). These metrics are used to evaluate the performance of the analysis task so that users can understand the accuracy and reliability of the analysis. The output of the evaluation agent is stored in JSON file format for easy feedback and archiving in subsequent tasks.

[0070] Visualization Agent:

[0071] The visualization agent receives the evaluation results and generates visual charts, such as ROC curves and scatter plots. Users can intuitively understand the effectiveness of the protein analysis through these images. The visualization agent then feeds back the final results (such as JPG files) to the chat manager and displays them to the user so that the user can understand the analysis results. See also Figure 3 This demonstrates the process by which the evaluation agent calculates metrics for the inference results and passes them to the visualization agent for image generation. The process mainly includes the calculation steps of the evaluation metrics and the generation of the final visualization results.

[0072] The ProtChat system employs a distributed multi-agent architecture, enabling efficient and smooth protein analysis tasks. Its design focuses on automatic task decomposition and dynamic module collaboration, integrating a natural language model (GPT-4) and a protein language model (MASSA) to automate and simplify protein analysis.

[0073] The key innovations and points to be protected in this invention are as follows:

[0074] The key innovations of this invention include:

[0075] Multi-agent collaborative architecture: The ProtChat system is based on a multi-agent architecture, which divides the protein analysis process into multiple independently executable modules, significantly improving the efficiency and accuracy of task processing.

[0076] Natural Language Task Command Parsing: The GPT-4 model is used to parse user commands in natural language, simplifying the interactive process of protein analysis and allowing users to operate the system directly through natural language commands.

[0077] Automated Multitask Analysis of Proteins: The reasoning agent, based on the MASSA model, supports the automated execution of various tasks such as protein attribute prediction, protein-drug interaction analysis, and protein-protein interaction, without requiring users to write additional scripts or perform data tuning.

[0078] The technical protection points of this invention include:

[0079] Division of labor and cooperation mechanism of multi-agent system: protect the key technical points of task decomposition, task allocation and task feedback in the multi-agent cooperation mechanism, including the interaction process of user agent, reasoning, evaluation and visualization agents.

[0080] Natural Language Command Parsing and Automatic Task Assignment Method: Protects the scheme of parsing natural language commands and automatically assigning protein analysis tasks through GPT-4, ensuring that users can operate the protein analysis process through natural language.

[0081] Multi-task analysis technology based on protein language model: Protect the MASSA model-based protein multi-task analysis technology, including application schemes in tasks such as protein property prediction, protein-drug interaction, and protein-protein interaction.

[0082] Compared with the prior art, the advantages of the present invention are:

[0083] Replacement of protein language model: The MASSA model in the ProtChat system can be replaced with other PLLMs with protein language representation capabilities to meet different task requirements or specific datasets.

[0084] Expanding task types: The system can add new protein analysis task modules, such as protein-nucleic acid interactions and protein structure prediction, to facilitate the application of the system in other bioinformatics tasks.

[0085] Adaptive optimization of agent collaboration: By introducing techniques such as reinforcement learning, the system can adaptively adjust the collaboration order and resource allocation of each agent according to the task complexity, so as to improve task processing efficiency.

[0086] Data interface expansion: The system can be expanded to include a data input interface, making it compatible with different formats of protein sequence and drug molecule data files, and adapting to more application scenarios.

[0087] The ProtChat system of this invention has been validated for its effectiveness and accuracy in protein analysis tasks through a series of experiments. Experimental results show that ProtChat can achieve efficient automation in multiple protein analysis tasks and significantly reduce the need for manual intervention. Specific results are as follows:

[0088] Protein attribute prediction: In the protein attribute prediction task, the ProtChat system demonstrates performance metrics comparable to existing state-of-the-art models, including accuracy and root mean square error (RMSE), and outperforms traditional manual preprocessing and single-model operations in prediction speed.

[0089] Protein-drug interaction prediction: Experiments validated the high accuracy of ProtChat in the protein-drug interaction prediction task (e.g., area under the ROC curve, AUC). The results show that the system can effectively predict the binding potential between proteins and drugs, significantly improving the efficiency of this task.

[0090] Protein-protein interaction prediction: In protein-protein interaction prediction, ProtChat achieves automated analysis and evaluation processes through multi-agent collaboration. The system demonstrates high accuracy and stability in both multi-classification and regression tasks, indicating its strong adaptability to handling complex bioinformatics tasks.

[0091] Overall, the experimental results demonstrate the effectiveness and feasibility of the ProtChat system in protein analysis tasks. Through automation and multi-task collaboration, ProtChat not only meets expectations in terms of accuracy but also significantly outperforms traditional methods in processing speed and user-friendliness, providing an efficient and practical solution for protein analysis and bioinformatics research.

[0092] The modified design or alternative solution of this invention is as follows:

[0093] Replace the Protein Language Model (PLLM):

[0094] The MASSA model used in the ProtChat system can be replaced with other protein language models suitable for protein data analysis (such as ESM, ProtTrans, etc.). This replacement can be selected based on the specific needs of the task or the characteristics of different datasets to optimize the system's task performance.

[0095] Multi-agent configuration adjustments:

[0096] The multi-agent configuration of the ProtChat system can be adjusted according to the needs of the actual task. For example, the number of reasoning, evaluation, or visualization agents can be increased or decreased based on the task complexity, or agent resources can be dynamically allocated through adaptive algorithms to improve task processing efficiency.

[0097] Model fine-tuning based on task requirements:

[0098] For specific tasks (such as predicting specific protein-drug interactions), the PLLM model used in ProtChat can be fine-tuned to enhance its performance on that task, thereby further improving the system's accuracy and applicability.

[0099] Compatibility with data from other fields:

[0100] The ProtChat system can expand its data interface to support other types of biological data analysis, such as the prediction of interactions between RNA, DNA, or other biomolecules. Through its modular design, the system can be compatible with different types of data, enabling a wider range of applications.

[0101] Other uses of the present invention:

[0102] Drug discovery and development:

[0103] ProtChat's protein-drug interaction prediction function can be used in the new drug development stage to quickly screen potential drug molecules and their target proteins, significantly improving the efficiency and accuracy of drug discovery and reducing R&D costs.

[0104] Biomarker discovery:

[0105] This system can be applied to the prediction and analysis of biomarkers. Through the automated prediction and evaluation of protein properties, it helps to discover disease-related biomarkers and provides support for precision medicine.

[0106] Protein design and optimization:

[0107] ProtChat can predict protein structure and function through multi-agent collaboration, providing data support for the design of artificial proteins. For example, it can be used to design proteins with specific functions or optimize the physical and chemical properties of proteins to meet the needs of industrial applications.

[0108] Education and training tools:

[0109] ProtChat's natural language interface and automated analysis capabilities make it an educational and training tool in the field of bioinformatics, facilitating protein analysis experiments for students and researchers and improving learning and research efficiency.

[0110] In summary, the ProtChat system is highly flexible and scalable, capable of adapting to various bioinformatics tasks, and demonstrates broad application potential in fields such as drug development, precision medicine, and education and training.

[0111] Example 2

[0112] With the development of large language models (LLMs), protein language models (PLLMs) have shown great potential in bioinformatics for processing protein data. However, existing PLLMs cannot effectively handle complex protein analysis tasks, especially when users without computational background need to perform protein analysis via natural language commands. Protein language models have significant shortcomings in task automation and the understanding and execution of task commands. Existing protein analysis workflows are complex, often requiring manual preprocessing, script building, and model tuning, leading to inefficiency. Therefore, there is an urgent need for an automated system that integrates natural language understanding and protein analysis functions to improve the efficiency of protein attribute prediction, protein-drug interaction, and protein-protein interaction tasks.

[0113] This invention aims to develop an agent-based automated protein property prediction system—ProtChat. By integrating the task planning capabilities of the large-scale language model GPT-4 with the analytical capabilities of a protein language model, it performs protein data parsing and prediction based on a general-purpose large language model and a general-purpose language model. Combining the advantages of the large-scale language model ChatGPT and the general-purpose language model, it provides life science researchers outside of computation with a framework for accurate protein physicochemical property prediction calculations using simple natural language. This invention can automatically execute protein analysis tasks, achieving automated processing from user commands to analysis results. ProtChat supports protein property prediction, protein-drug interaction, and protein-protein interaction analysis, and automatically generates visualization results, thereby significantly improving protein analysis efficiency and lowering the user threshold.

[0114] According to an embodiment of the present invention, an automated protein property prediction system based on an intelligent agent is provided, see [link to relevant documentation]. Figure 3 ,include:

[0115] User agent 201 is used to receive protein analysis instructions input by the user;

[0116] Chat management agent 202 is used to parse protein analysis instructions, identify the core requirements of protein analysis instructions, and determine the required analysis task type.

[0117] The reasoning agent 203 is used to perform protein data analysis according to the analysis task type, and completes the parsing and prediction of protein data based on a general language model, and generates prediction results.

[0118] The agent-based automated protein property prediction system in this embodiment of the invention can automatically execute protein analysis tasks, realize automated processing from user instructions to analysis results, thereby significantly improving protein analysis efficiency and reducing the user threshold.

[0119] The core technical solution of this invention includes the following modules:

[0120] User Proxy module: This module is responsible for receiving user input commands and passing them to the Chat Manager to generate task plans.

[0121] Inference Agent: Responsible for performing protein analysis tasks, calling the MASSA model to process protein data, and supporting various tasks such as protein attribute prediction, protein-drug interaction analysis, and protein-protein interaction analysis.

[0122] Evaluation Agent: Analyzes and evaluates the inference results, calculates and analyzes task metrics (such as accuracy, AUC, etc.), and feeds the results back to the "Chat Manager".

[0123] Visualization Agent: Visualizes the analysis results of a task and generates user-friendly results displays such as images.

[0124] ProtChat's multi-agent system can dynamically allocate and coordinate tasks, enabling command-driven automated protein analysis. Through simple user command interaction, the system can automatically identify the required tools and task flows, thereby greatly simplifying the complexity and reducing the barrier to entry for protein analysis.

[0125] The technical solution of the present invention is described in detail below:

[0126] The ProtChat system of this invention automates protein analysis tasks through multiple collaborative intelligent agent modules. See also... Figure 2 This diagram illustrates the overall architecture of the ProtChat system, including the relationships between the user agent, chat manager, inference agent, evaluation agent, and visualization agent. The diagram clearly shows the functional division of each module and their collaborative methods. An example of a user interface inputting tasks via natural language commands, and the visual results provided by the system after task completion, demonstrate ProtChat's user-friendly interface. The following are the detailed technical solutions for each agent module in the ProtChat system:

[0127] User Proxy module:

[0128] This module serves as the interface between the user and the system, receiving protein analysis instructions from the user and supporting instruction input in natural language. The user agent passes the instructions to the "Chat Manager" module for task parsing, identifies the core requirements of the instructions, and determines the required analysis task type (such as protein property prediction, protein-drug interaction, protein-protein interaction, etc.).

[0129] Chat Manager:

[0130] The chat manager receives instructions from the user agent and assigns tasks to the appropriate agents. Its core function is to break down protein analysis tasks into multiple sub-tasks and coordinate the sequential execution of these sub-tasks by the agents. The chat manager leverages GPT-4's natural language understanding capabilities to parse complex task requirements, thereby ensuring the accuracy of task decomposition and execution.

[0131] Inference Agent:

[0132] The inference agent is responsible for performing the analysis of protein data, parsing and predicting protein data based on a general language model. The inference agent includes several customized analysis function modules, such as "Analyze_Protein_Property", "Analyze_Protein_Drug_Interaction", and "Analyze_Protein_Protein_Interaction". These modules can automatically call appropriate data processing methods for different protein analysis tasks, generate prediction results, and return them to the chat manager.

[0133] Evaluation Agent:

[0134] The evaluation agent is responsible for analyzing the predictions generated by the inference agent and calculating task metrics (such as accuracy, ROC curve, RMSE, etc.). These metrics are used to evaluate the performance of the analysis task so that users can understand the accuracy and reliability of the analysis. The output of the evaluation agent is stored in JSON file format for easy feedback and archiving in subsequent tasks.

[0135] Visualization Agent:

[0136] The visualization agent receives the evaluation results and generates visual charts, such as ROC curves and scatter plots. Users can intuitively understand the effectiveness of the protein analysis through these images. The visualization agent then feeds back the final results (such as JPG files) to the chat manager and displays them to the user so that the user can understand the analysis results. See also Figure 3 This demonstrates the process by which the evaluation agent calculates metrics for the inference results and passes them to the visualization agent for image generation. The process mainly includes the calculation steps of the evaluation metrics and the generation of the final visualization results.

[0137] The ProtChat system employs a distributed multi-agent architecture, enabling efficient and smooth protein analysis tasks. Its design focuses on automatic task decomposition and dynamic module collaboration, integrating a natural language model (GPT-4) and a protein language model (MASSA) to automate and simplify protein analysis.

[0138] The key innovations and points to be protected in this invention are as follows:

[0139] The key innovations of this invention include:

[0140] Multi-agent collaborative architecture: The ProtChat system is based on a multi-agent architecture, which divides the protein analysis process into multiple independently executable modules, significantly improving the efficiency and accuracy of task processing.

[0141] Natural Language Task Command Parsing: The GPT-4 model is used to parse user commands in natural language, simplifying the interactive process of protein analysis and allowing users to operate the system directly through natural language commands.

[0142] Automated Multitask Analysis of Proteins: The reasoning agent, based on the MASSA model, supports the automated execution of various tasks such as protein attribute prediction, protein-drug interaction analysis, and protein-protein interaction, without requiring users to write additional scripts or perform data tuning.

[0143] The technical protection points of this invention include:

[0144] Division of labor and cooperation mechanism of multi-agent system: protect the key technical points of task decomposition, task allocation and task feedback in the multi-agent cooperation mechanism, including the interaction process of user agent, reasoning, evaluation and visualization agents.

[0145] Natural Language Command Parsing and Automatic Task Assignment Method: Protects the scheme of parsing natural language commands and automatically assigning protein analysis tasks through GPT-4, ensuring that users can operate the protein analysis process through natural language.

[0146] Multi-task analysis technology based on protein language model: Protect the MASSA model-based protein multi-task analysis technology, including application schemes in tasks such as protein property prediction, protein-drug interaction, and protein-protein interaction.

[0147] Compared with the prior art, the advantages of the present invention are:

[0148] Replacement of protein language model: The MASSA model in the ProtChat system can be replaced with other PLLMs with protein language representation capabilities to meet different task requirements or specific datasets.

[0149] Expanding task types: The system can add new protein analysis task modules, such as protein-nucleic acid interactions and protein structure prediction, to facilitate the application of the system in other bioinformatics tasks.

[0150] Adaptive optimization of agent collaboration: By introducing techniques such as reinforcement learning, the system can adaptively adjust the collaboration order and resource allocation of each agent according to the task complexity, so as to improve task processing efficiency.

[0151] Data interface expansion: The system can be expanded to include a data input interface, making it compatible with different formats of protein sequence and drug molecule data files, and adapting to more application scenarios.

[0152] The ProtChat system of this invention has been validated for its effectiveness and accuracy in protein analysis tasks through a series of experiments. Experimental results show that ProtChat can achieve efficient automation in multiple protein analysis tasks and significantly reduce the need for manual intervention. Specific results are as follows:

[0153] Protein attribute prediction: In the protein attribute prediction task, the ProtChat system demonstrates performance metrics comparable to existing state-of-the-art models, including accuracy and root mean square error (RMSE), and outperforms traditional manual preprocessing and single-model operations in prediction speed.

[0154] Protein-drug interaction prediction: Experiments validated the high accuracy of ProtChat in the protein-drug interaction prediction task (e.g., area under the ROC curve, AUC). The results show that the system can effectively predict the binding potential between proteins and drugs, significantly improving the efficiency of this task.

[0155] Protein-protein interaction prediction: In protein-protein interaction prediction, ProtChat achieves automated analysis and evaluation processes through multi-agent collaboration. The system demonstrates high accuracy and stability in both multi-classification and regression tasks, indicating its strong adaptability to handling complex bioinformatics tasks.

[0156] Overall, the experimental results demonstrate the effectiveness and feasibility of the ProtChat system in protein analysis tasks. Through automation and multi-task collaboration, ProtChat not only meets expectations in terms of accuracy but also significantly outperforms traditional methods in processing speed and user-friendliness, providing an efficient and practical solution for protein analysis and bioinformatics research.

[0157] The modified design or alternative solution of this invention is as follows:

[0158] Replace the Protein Language Model (PLLM):

[0159] The MASSA model used in the ProtChat system can be replaced with other protein language models suitable for protein data analysis (such as ESM, ProtTrans, etc.). This replacement can be selected based on the specific needs of the task or the characteristics of different datasets to optimize the system's task performance.

[0160] Multi-agent configuration adjustments:

[0161] The multi-agent configuration of the ProtChat system can be adjusted according to the needs of the actual task. For example, the number of reasoning, evaluation, or visualization agents can be increased or decreased based on the task complexity, or agent resources can be dynamically allocated through adaptive algorithms to improve task processing efficiency.

[0162] Model fine-tuning based on task requirements:

[0163] For specific tasks (such as predicting specific protein-drug interactions), the PLLM model used in ProtChat can be fine-tuned to enhance its performance on that task, thereby further improving the system's accuracy and applicability.

[0164] Compatibility with data from other fields:

[0165] The ProtChat system can expand its data interface to support other types of biological data analysis, such as the prediction of interactions between RNA, DNA, or other biomolecules. Through its modular design, the system can be compatible with different types of data, enabling a wider range of applications.

[0166] Other uses of the present invention:

[0167] Drug discovery and development:

[0168] ProtChat's protein-drug interaction prediction function can be used in the new drug development stage to quickly screen potential drug molecules and their target proteins, significantly improving the efficiency and accuracy of drug discovery and reducing R&D costs.

[0169] Biomarker discovery:

[0170] This system can be applied to the prediction and analysis of biomarkers. Through the automated prediction and evaluation of protein properties, it helps to discover disease-related biomarkers and provides support for precision medicine.

[0171] Protein design and optimization:

[0172] ProtChat can predict protein structure and function through multi-agent collaboration, providing data support for the design of artificial proteins. For example, it can be used to design proteins with specific functions or optimize the physical and chemical properties of proteins to meet the needs of industrial applications.

[0173] Education and training tools:

[0174] ProtChat's natural language interface and automated analysis capabilities make it an educational and training tool in the field of bioinformatics, facilitating protein analysis experiments for students and researchers and improving learning and research efficiency.

[0175] In summary, the ProtChat system is highly flexible and scalable, capable of adapting to various bioinformatics tasks, and demonstrates broad application potential in fields such as drug development, precision medicine, and education and training.

[0176] Example 3

[0177] A storage medium storing program files capable of implementing any of the above-mentioned agent-based automated protein property prediction methods.

[0178] Example 4

[0179] A processor for running a program, wherein the program executes any of the above-mentioned agent-based automated protein property prediction methods during runtime.

[0180] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0181] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0182] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units 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 of units or modules may be electrical or other forms.

[0183] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0184] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0185] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0186] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An automated protein property prediction method based on intelligent agents, characterized in that, Includes the following steps: S101: Receives the protein analysis task instruction input by the user and decomposes the analysis task instruction; S102: Perform task parsing on the decomposed analysis task instructions, access the custom functions for protein understanding tasks, identify the core requirements of the protein analysis task instructions, and determine the required analysis task type. S103: According to the analysis task type, access the custom function, complete the parsing and prediction of protein data based on the general language model, and generate inference results.

2. The automated protein property prediction method based on intelligent agents according to claim 1, characterized in that, The method further includes: Analyze the generated inference results, access custom functions, and calculate task metrics based on the inference results.

3. The automated protein property prediction method based on intelligent agents according to claim 2, characterized in that, The method further includes: Access customizable features and perform visual operations based on task metrics.

4. The automated protein property prediction method based on intelligent agents according to claim 1, characterized in that, In step S102, the analysis task types include: protein property prediction, protein-drug interaction, and protein-protein interaction.

5. The automated protein property prediction method based on intelligent agents according to claim 1, characterized in that, In step S102, the protein analysis task is decomposed into multiple sub-tasks and executed in a coordinated manner. The complex task requirements are parsed by calling the natural language understanding capabilities of GPT-4, ensuring the accuracy of task decomposition and execution.

6. The automated protein property prediction method based on intelligent agents according to claim 1, characterized in that, The general language model includes the general large language model and / or the general protein language model.

7. An automated protein property prediction system based on intelligent agents, characterized in that, include: The user intelligent agent is used to receive protein analysis task instructions input by the user and decompose the analysis task instructions. The chat management agent is used to parse the decomposed analysis task instructions, access custom functions for protein understanding tasks, identify the core requirements of protein analysis task instructions, and determine the required analysis task type. The reasoning agent is used to access custom functions according to the type of analysis task, to parse and predict protein data based on a general language model, and to generate reasoning results.

8. The predictive analysis system according to claim 7, characterized in that, The system also includes: Evaluate the agent to analyze the generated inference results, access custom functions, and calculate task metrics based on the inference results.

9. The predictive analysis system according to claim 8, characterized in that, The system also includes: Visual agents are used to access custom functions and perform visual operations based on task metrics.

10. The predictive analysis system according to claim 7, characterized in that, The reasoning agent includes multiple customized analysis function modules. These modules automatically call appropriate data processing methods to generate prediction results for different protein analysis tasks.

Images