System and method for processing molecular data of a drug based on an ai agent

Abstract

The invention relates to a system and a method for processing molecular data of a drug based on an AI agent. The method comprises the following steps: constructing an AI agent, wherein the AI agent comprises a large language model, a cue word module, a tool library module, a memory module and an output parser; inputting the task description of the natural language of the user and the molecular data of the medicine to be processed into the constructed AI agent to understand the intention of the user and analyze the property of the molecular data; according to the intention of the user and the property of the molecular data, performing autonomous reasoning by utilizing priori knowledge included in a large language model and a cue word module to generate a global operation sequence; executing self-consistent circulation on the global operation sequence so as to process molecular data of the medicine by utilizing a tool library module; and outputting a final processing result of the molecular data of the drug and presenting the final processing result to a user.

Description

Technical Field

[0001] The present invention relates to general artificial intelligence and data processing, and in particular to a system and method for processing molecular data of drugs based on artificial intelligence (AI) agents. Background Art

[0002] New drug development is a challenging task, which requires not only considering the biological activity and selectivity of the drug for the target, but also the absorption, distribution, metabolism, excretion and toxicity (ADMET) of the drug in the human body. After years of new drug development, the field of small molecule innovative drugs has accumulated a large amount of data covering various molecular structures, activities and ADMET properties.

[0003] The molecular data of drugs refers to the information on the chemical structure, physical properties, biological activity, etc. of the compound. For example, the chemical structure provides the two-dimensional and three-dimensional conformations of the compound; the physical properties include parameters such as melting point, boiling point, solubility, etc.; the biological activity data records the effect of the compound on a specific biological target, as well as pharmacokinetic and toxicity data. All of this information together constitutes a comprehensive and complex data set, providing a solid foundation for drug design and screening.

[0004] On the one hand, by studying the quantitative structure-activity relationship (QSAR) in this type of molecular big data, new drug developers can find suitable leads or lead compounds. On the other hand, by analyzing the molecular data of massive drugs, researchers can adjust the structure of existing compounds and optimize ADMET properties. Therefore, molecular big data will become an important boost to drug development and is expected to become the key to improving the efficiency and success rate of drug development.

[0005] Considering the importance of molecular big data, how to process such data has also become a focus of attention in recent years. The molecular data processing of drugs includes modules such as mining, cleaning, database import and export, search, conversion and analysis of molecular information. With the development of artificial intelligence and big data technology, the molecular data processing modules of these drugs mentioned above can be implemented through deep neural networks or other advanced algorithms. For example, the mining of molecular information can be performed through the AI ​​model of joint table extraction and content recognition to extract information from patents or documents; Pandas in the Python library can be used to perform operations such as data reading and writing, normalization, deduplication, merging and splitting to achieve data cleaning; SQL language can realize the rapid import and export of drug molecular data.

[0006] Although molecular big data has promoted the development of new drugs, its data scale is still far behind the data volume of computer vision and natural language processing, and the data scale limits the application of various generative models and discriminative models in new drug development. Wet laboratory R&D personnel will generate a large amount of valuable data in the course of their work. Wet laboratory R&D personnel usually refer to technicians who deal directly with liquid samples, chemical reagents or biological materials in a laboratory environment. Wet laboratory R&D personnel are mainly responsible for experimental research and data collection in the fields of biology, biochemistry, chemistry, pharmacology, etc.

[0007] If wet lab researchers can quickly process this type of data, the scale of molecular data for new drug development can be rapidly increased, further promoting the efficiency and digital transformation of new drug development. However, most wet lab researchers do not have backgrounds in AI, programming, and databases, which prevents them from flexibly applying the molecular data processing technology for the above drugs. Therefore, it is necessary to strengthen interdisciplinary cooperation, improve the quality, scale, and processing capabilities of molecular data for drugs, and thus promote the efficiency of new drug development.

[0008] The above description of the background technology is only for facilitating an in-depth understanding of the technical solution of the present invention (such as the technical means used, the technical problems solved and the technical effects produced), and should not be regarded as an admission or any form of implication that the information constitutes prior art already known to those skilled in the art. Summary of the invention

[0009] In view of the defects existing in the prior art, the present invention proposes a novel system and method for processing molecular data of drugs based on AI agents, wherein the AI ​​agent includes a large language model, prompt words, a memory module and a tool library module. After the user inputs the task description of the molecular data processing of drugs in natural language into this system and uploads any form of molecular representation, the AI ​​agent can analyze user needs, make plans, call internal tools, execute plans based on the large language model, and use carefully designed prompt word engineering to analyze the effect of task execution, and finally realize end-to-end diversified processing of molecular data of drugs, including mining, cleaning, database import and export, search, conversion and analysis.

[0010] According to an embodiment of the present invention, a method for processing molecular data of drugs based on an AI agent is provided, which includes: constructing an AI agent, wherein the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module and an output parser; inputting a user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data; performing autonomous reasoning based on the user's intention and the properties of the molecular data using the large language model and the prior knowledge included in the prompt word module to generate a global operation sequence; executing a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library module; and outputting the final processing result of the molecular data of the drug and presenting it to the user.

[0011] Preferably, the large language model is a pre-trained deep learning model, which includes a non-open source large language model and / or an open source large language model, and the large language model is configured to understand the user's natural language and analyze the user's intention; the prompt word module is configured to assist the large language model in understanding and analyzing the user's intention and determine the molecular data processing task in combination with the memory module; the tool library module is configured to integrate tools for processing the molecular data of drugs, and includes a text description of the tools for processing the molecular data of drugs; the memory module is configured to store intermediate state information related to subtasks and historical information of user interactions, so as to enhance the perception, decision-making and action of the molecular data processing task; the output parser is configured to post-process the output generated by the large language model to ensure that the result conforms to the expected format or standard.

[0012] Preferably, executing a self-consistent loop on the global operation sequence includes: a thinking sub-step, which determines the task type and the execution target of the sub-task according to the global operation sequence, and the task type includes a molecular data processing task of a drug completed by a single sub-task and a molecular data processing task of a drug completed by multiple sub-tasks; a planning sub-step, which determines the action sequence of the sub-task in combination with the text description of the tool in the tool library module; a critique sub-step, which evaluates the action sequence of the sub-task, determines potential problems in combination with the information stored in the memory module, and corrects the action sequence of the sub-task; an action sub-step, which uses the tool library module to execute the corrected action sequence of the sub-task; and an observation sub-step, which determines the progress of the task based on the results of the thinking sub-step and the results of the action sub-step, combined with the historical information of the memory module and the prompt word module and the text description of the tool in the tool library module.

[0013] Preferably, executing a self-consistent loop on the global operation sequence further includes: when the task type determined by the thinking sub-step is a molecular data processing task of a drug completed by a single sub-task, the observing sub-step outputs the final processing result of the molecular data of the drug; when the task type determined by the thinking sub-step is a molecular data processing task of a drug completed by multiple sub-tasks, automatically entering the loop sub-step until the multiple sub-tasks are completed and the final processing result of the molecular data of the drug is output; the loop sub-step is configured to continue executing a new sub-task based on the result of the thinking sub-step and the result of the observing sub-step, wherein executing the new sub-task includes a thinking sub-step, a planning sub-step, a criticizing sub-step, an acting sub-step and an observing sub-step.

[0014] Preferably, the prompt word module is configured as follows: including a description of the task of processing the molecular data of the drug; including a task description of processing the molecular data of the drug and prior knowledge of the calling tool flow of complex tasks; using hierarchical prompt words to guide the large language model to learn step by step.

[0015] Preferably, the tool library module includes an application programming interface of the underlying algorithm; the tool library module further includes at least one of a data mining tool, a data cleaning tool, a database import and export tool, a data search tool, a data conversion tool and a data analysis tool for processing the molecular data of drugs.

[0016] Preferably, the memory module includes a long-term memory component and a short-term memory component; the long-term memory component is configured to store data related to user interaction history; the short-term memory component is configured to store intermediate state information related to subtasks, which is used to identify potential problems in the critical sub-step of the self-consistent loop.

[0017] Preferably, the user's natural language task description includes at least one of the operations of extracting, storing, mining, predicting, visualizing and searching for molecular data of drugs; the molecular data includes molecular structure data and files containing molecular structure data and molecular property data; the molecular structure data includes at least one of one-dimensional representation, two-dimensional representation and three-dimensional representation of the molecular structure, and the molecular property data includes at least one of the properties of activity, absorption, distribution, metabolism, excretion and toxicity of the drug; wherein the one-dimensional representation of the molecular structure includes a file in SMILES format; the two-dimensional representation of the molecular structure includes a molecular picture in JPG or PNG or a molecular file of atoms and chemical bond positions in MOL or SDF; the three-dimensional representation of the molecular data includes molecular files in PDB, CIF and MOL2; the files containing molecular structure data and molecular property data include list-form files and database files, the list-form files include CSV and XLSX files, and the database files include SQL class files.

[0018] Preferably, the final processing result of the AI ​​agent includes: at least one of a one-dimensional representation, a two-dimensional representation and a three-dimensional representation of a molecular structure, a file containing molecular structure data and molecular property data, and a QSAR model of a drug molecule.

[0019] According to an embodiment of the present invention, a system for processing molecular data of drugs based on an AI agent is provided, which includes: a construction module, which is configured to construct an AI agent, and the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module and an output parser; a user intention understanding module, which is configured to input the user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data; an autonomous reasoning module, which is configured to perform autonomous reasoning based on the user's intention and the properties of the molecular data using the large language model and the prior knowledge included in the prompt word module to generate a global operation sequence; a self-consistent loop module, which is configured to perform a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library module; and an output module, which is configured to output the final processing result of the molecular data of the drug and present it to the user.

[0020] According to an embodiment of the present invention, there is provided a computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions, when executed by a processor, implement the following steps: construct an AI agent, wherein the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module, and an output parser; input a user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data; perform autonomous reasoning based on the user's intention and the properties of the molecular data using the large language model and the prior knowledge included in the prompt word module to generate a global operation sequence; execute a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library module; output the final processing result of the drug's molecular data and present it to the user.

[0021] The present invention adopts the above technical solution, which has the following beneficial effects:

[0022] 1. Automation and efficiency: By using AI agents to automatically process and analyze a large amount of drug molecular data, the need for human involvement is reduced, and the speed and efficiency of data processing are improved. This automated processing not only saves time, but also helps reduce human errors, ensures the consistency and reliability of data analysis, and meets the growing demand of wet laboratory R&D personnel for drug molecular data processing.

[0023] 2. Versatility: By building the first AI agent-specific molecular data tool library, various molecular data and files can be processed. The molecular data tool library contains relevant algorithms and software for processing various molecular data and files, which promotes the coordination of the large language model as a decision-making module and the tool library in the process of global and local reasoning and decision-making of AI agents. The molecular data tool library not only promotes the smooth operation of various molecular data processing processes, but also ensures the high compatibility of various molecular data types as input and output.

[0024] 3. Strong learning ability: By leveraging machine learning and deep learning technologies, it can learn from data and continuously optimize its analysis algorithms.

[0025] 4. Interdisciplinary integration: Combining software and algorithm development with the needs of biomedical researchers removes the barriers to data processing between software and algorithm developers and wet experiment researchers, breaking the boundaries between traditional disciplines. This promotes cooperation between experts in different fields, allowing computer technology and biological knowledge to advance together.

[0026] 5. Scalability and modularity: With the continuous development of AI technology, it allows easy addition of new functional modules and updating of existing algorithms, ensuring the long-term applicability and flexibility of the system.

[0027] 6. Promote drug discovery: By quickly and accurately processing the molecular data of drugs, it promotes the continuous progress of molecular big data in empowering new drug research and development.

[0028] The system and method for processing molecular data of drugs based on AI agents according to the implementation scheme of the present invention can enable AI agents to learn to use algorithms and software, meet the growing demand of wet experiment researchers for molecular data processing of drugs, remove the barriers to data processing between software and algorithm developers and wet experiment researchers, and the system always maintains scalability, promoting the continuous progress of molecular big data-enabled new drug research and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The exemplary embodiments of the present invention will be described in more detail below in conjunction with the accompanying drawings. For the sake of clarity, the same components in different drawings are shown with the same reference numerals. It should be noted that the drawings are only for illustration and are not necessarily drawn to scale. In these drawings:

[0030] Figure 1 is a flowchart illustrating a method for processing molecular data of drugs based on an AI agent according to an embodiment of the present invention.

[0031] Figure 2 It is a flowchart showing that a method for processing molecular data of drugs based on an AI agent according to an embodiment of the present invention performs a self-consistent loop on a generated global operation sequence.

[0032] Figure 3 The method for processing the molecular data of drugs based on an AI agent according to an embodiment of the present invention is shown as a processing process of the task of processing the molecular data of drugs that can be completed through a single subtask.

[0033] Figure 4 The SMILES representation of the original drug molecular data is shown.

[0034] Figure 5 The figure shows the molecular data of the drug after being processed by executing a self-consistent cycle using the method for processing the molecular data of the drug based on an AI agent according to the present invention.

[0035] Figure 6 The figure shows the processing process of the task of processing the molecular data of drugs which can be completed through multiple subtasks according to the method of processing the molecular data of drugs based on AI agents according to an embodiment of the present invention.

[0036] Figure 7 is a block diagram illustrating a system for processing molecular data of drugs based on an AI agent according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The implementation scheme of the present invention is described in detail below. This implementation scheme is implemented on the premise of the technical scheme of the present invention, and a detailed implementation method and specific operation process are given, but the protection scope of the present invention is not limited to the implementation scheme described below.

[0038] Figure 1 is a flow chart showing a method for processing drug molecular data based on an AI agent according to an embodiment of the present invention. Figure 1 According to an embodiment of the present invention, the method for processing drug molecular data based on an AI agent may include the following steps:

[0039] Step S100, constructing an AI agent, wherein the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module and an output parser.

[0040] The large language model can be configured to understand natural language; the prompt word module can be configured to contain prior knowledge for mining, cleaning, database import and export, search, conversion and analysis of drug molecular data, to assist the large language model in understanding and analyzing user intentions, and to determine subsequent molecular data processing tasks in combination with the memory module; the tool library module is a collection of various tools (algorithms, software) for drug molecular data processing with added tool descriptions; the memory module is a structural unit for enhancing the perception, decision-making and action of molecular data processing tasks by storing historical information of task execution and user interaction; the output parser can be configured to post-process the output generated by the large language model.

[0041] Step S200, inputting the user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data.

[0042] Step S300 , based on the user intention and the properties of the molecular data, autonomous reasoning is performed using the prior knowledge included in the large language model and the prompt word module to generate a global operation sequence.

[0043] Step S400, execute a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library module. According to the global operation sequence inferred autonomously, the AI ​​agent automatically selects and calls chemical information processing tools or algorithms to perform specific data processing tasks. The AI ​​agent can repeatedly execute these tools, adjust and optimize parameters based on intermediate results, until the expected analysis accuracy is achieved or the user's specific instructions are completed.

[0044] Step S500: output the final processing result of the drug's molecular data and present it to the user.

[0045] The AI ​​agent can output the final processing results to the user, which can be a sorted data set, image file, report form or model file, etc. At the same time, the AI ​​agent can also provide explanations of the results to help users understand the results of complex models or data analysis.

[0046] The AI ​​agent according to the embodiment of the present invention is a system based on artificial intelligence technology that can understand natural language, process the molecular data of drugs and make intelligent responses or actions, thereby automatically performing tasks or making decisions in its environment. The self-consistent loop of the AI ​​agent can not only guide the agent to further understand the task requirements, but also further plan and execute specific operations. In addition, the self-consistent loop allows the AI ​​agent to perform multiple rounds of reasoning and action, which is crucial for handling complex tasks. Finally, by outputting the final processing result, the AI ​​agent completes its task process and provides the user with the required information or services.

[0047] The following is a detailed description of the components of the AI ​​agent according to the implementation scheme of the present invention.

[0048] The AI ​​agent according to an embodiment of the present invention may include a large language model, a prompt word module, a tool library module, a memory module and an output parser.

[0049] The large language model of the AI ​​agent according to the embodiment of the present invention is a pre-trained deep learning model with the ability to process and understand natural language. The large language model can be built based on different architectures and algorithms, aiming to learn the laws and patterns of language from a large amount of text data. According to the embodiment of the present invention, the large language model can use non-open source large language models such as GPT, Claude, Gemini, or open source large language models such as Llama, Mistral, and Qwen. GPT is a generative pre-trained transformer model developed by OpenAI that can generate coherent natural language text and is used for a variety of language processing tasks. Claude is a large language model developed by Anthropic that focuses on providing creative and educational answers. Gemini is a large language model launched by DeepMind that aims to understand and generate complex natural language text. Llama is an open source large language model launched by Facebook AI that aims to promote scalable and efficient model training. Mistral is an open source large language model launched by Mistral AI, known for its multilingual capabilities and openness. Qwen is an open source large language model developed by Alibaba Cloud that focuses on science and technology-related question-answering tasks. According to an embodiment of the present invention, the large language model is not limited to the above-mentioned model, and may also be replaced by other models that can realize understanding of the input natural language.

[0050] In natural language processing (NLP), prompts are a key technology that guides AI agents to perform specific tasks by providing them with explicit guidance or contextual information. Prompts can be simple text snippets or more complex signals, and their purpose is to help large language models understand the user's intent more accurately and produce appropriate outputs.

[0051] The prompt word module of the AI ​​agent according to the embodiment of the present invention may include prior knowledge about the processing methods such as mining, cleaning, database import and export, search, conversion and analysis of the molecular data of the drug. The prompt words serve as a supplement to the natural language input of human beings, play a prompting role, and enhance the understanding ability of the large language model to the natural language input of human beings, so as to accurately analyze the user's intention.

[0052] The setting of the prompt word module according to the implementation scheme of the present invention is mainly aimed at the following three aspects:

[0053] (1) Clarity: When constructing prompts, it is necessary to accurately describe the task. For example, use a statement such as "extract data with CDK2 target molecular activity less than 50nM" instead of vaguely proposing "extract data with high CDK2 target biological activity." This clear statement helps provide a clear goal for the model, thereby better guiding the learning process and evaluating the performance of the model.

[0054] (2) Relevance: The constructed prompt words include prior knowledge of drug molecular data mining, cleaning, database import and export, search, conversion, and analysis, which helps ensure that the features that the model focuses on are consistent with those of medicinal chemistry. By incorporating professional knowledge into the prompt words, the performance of the model in specific fields can be improved.

[0055] (3) Strategic: Use hierarchical prompt words to guide the large language model to learn step by step. Start with the overall structure and then gradually refine to the recognition of micro-organizational features. This approach imitates the process of human learning and problem solving, and can help the model better understand the internal structure and logical relationship of complex data.

[0056] In short, the prompt word module plays a vital role in the application of AI agents. By ensuring the clarity of tasks, improving task relevance, and adopting strategic learning methods, the learning efficiency of the model and its ability to handle problems in specific fields can be greatly improved.

[0057] The tool library module of the AI ​​agent according to the implementation scheme of the present invention is an integration of various algorithms and software for drug molecule data processing, and the added tools are configured to provide important references when the AI ​​agent processes molecular data, so that the tool library module can be flexibly called to perform tasks such as molecular data mining, cleaning, database import and export, search, conversion and analysis.

[0058] When processing the molecular data of a drug, it is essential to select appropriate tools and algorithms to obtain accurate analysis results. The tool library for processing the molecular data of a drug according to an embodiment of the present invention may include data mining tools, data cleaning tools, database import and export tools, data search tools, data conversion tools, data analysis tools, etc., for processing the molecular data of a drug.

[0059] The tool library of the AI ​​agent according to the implementation scheme of the present invention includes not only the application programming interface (API) of the underlying algorithm and software, but also text descriptions of various tools, which improves the compatibility of the large language model with tool algorithms and software calls during the reasoning decision-making process. Furthermore, after understanding the task, the large language model can combine the text descriptions of various tools to efficiently determine the tools used in the reasoning process and the input of each step.

[0060] Data mining tools are used to extract important information from patents and papers, including compound structure, number, biological activity and ADMET properties. This type of information generally exists in the above-mentioned documents in tabular form. By combining table recognition and extraction and OCR technology, the above information can be extracted and mined.

[0061] Data cleaning tools can use mathematical algorithms to perform data sorting tasks such as deduplication, averaging, and removing invalid SMILES.

[0062] The database import and export tool can export the molecular structure, target information, biological activity, ADMET properties and other information in the database by using SQL language, or store the above information in the corresponding database.

[0063] The data search tool searches for molecules that meet the criteria in the database using SQL language based on a specific molecular structure (which can be a complete molecule or a substructure), and obtains information on the corresponding activity and ADMET properties of the molecule.

[0064] Data conversion tools use various algorithms, such as RDKit and DECIMER, to achieve conversion of molecular files between different formats. The files include CSV (SMILES or SMARTS to represent molecules), CIF, PDB, images (PNG or JPG), etc.

[0065] Data analysis tools can quantify the relationship between molecular structure and properties and establish various relationships. Data analysis tools can be used as judgment models to predict the properties of unknown molecules.

[0066] The function of the memory module of the AI ​​agent according to the implementation scheme of the present invention is to ensure the human-computer interaction of the AI ​​agent in the process of molecular data processing. The memory module may include a long-term memory component and a short-term memory component. The long-term memory component stores data related to the user interaction history, which is used to guide the implementation of future molecular data processing tasks; the short-term memory component stores the intermediate state information related to the subtask, which plays an important role in the critical substep of the self-consistent cycle to be described later. Based on the information stored in it, potential problems can be discovered in the process of executing the critical substep, and solutions can be provided for problems that arise in the execution of the subtask.

[0067] The role of the output parser of the AI ​​agent according to the embodiment of the present invention is to post-process the output generated by the large language model to ensure that the result conforms to the expected format or standard. For example, if the task requires that the molecular data of the drug be saved to a specific path, the output parser can ensure that these features are correctly formatted and stored in the specified location, which can improve the structure of the output of the large language model. In drug research and development, this may include converting the chemical structure predicted by the model into a standard chemical file format, for example, different representations of molecular data (such as SMILES, MOL, SDF, PDB, CIF, MOL2, etc.), CSV or XLSX files containing molecular structures and properties, updated drug molecule SQL databases, QSAR models, etc.

[0068] The output parser of the AI ​​agent according to the implementation scheme of the present invention can automatically process, convert and store pre-calculated data, reduce manual intervention and improve efficiency. Not only is it limited to improving the accuracy and efficiency of data processing, it can also ensure the consistency and availability of data formats, so that scientific researchers can devote more energy to scientific research itself rather than tedious data management tasks. In addition, by utilizing various professional databases, the drug development process can be significantly accelerated. Combined with efficient output parsing technology and a wide range of professional database resources, AI technology can not only accelerate the process of drug development, but also improve the success rate and quality of research and development.

[0069] The following is a detailed description of each step of the method for processing molecular data of drugs based on an AI agent according to an embodiment of the present invention.

[0070] In step S200, the user's natural language task description includes at least one of operations for extracting, storing, mining, predicting, visualizing, and searching for molecular data of drugs. The natural language may include at least one of text, video, and audio.

[0071] The AI ​​agent according to the implementation scheme of the present invention can understand and analyze the user's intention described by the user in natural language through a large language model. For example, the natural language description is "perform QSAR modeling on the compounds of the MAT2A target in the ChEMBL database", and the user intention is determined as: extracting the compound structure and activity information of the MAT2A target in the ChEMBL database, and establishing a QSAR model based on the above information.

[0072] Molecular data may include molecular structure data and files containing molecular structure data and molecular property data. Molecular structure data may be a one-dimensional representation of the molecular structure in the form of a string, such as a file in SMILES format; it may also be a two-dimensional representation of the molecular structure, such as a molecular picture (such as JPG, PNG, etc.) or a professional molecular file representing the position of atoms and chemical bonds (such as MOL, SDF files, etc.); it may also be a three-dimensional representation of the molecular structure, such as PDB, CIF, and MOL2 files. Molecular property data may include the properties of the drug in terms of activity, absorption, distribution, metabolism, excretion, and toxicity. Files containing molecular structure data and molecular property data include list-format files and database files. List-format files include CSV and XLSX files, and database files include SQL-class files.

[0073] Single molecule data can be saved in the above-mentioned forms, while complex molecular data (e.g., data containing multiple molecules or containing both structure and property information) can be saved as structured data files (e.g., CSV, XLSX, SQL type databases), or unstructured data files (e.g., DOC or PDF files describing the molecular structure and properties).

[0074] The AI ​​agent according to the embodiment of the present invention can also analyze the properties of the input molecular data to identify the molecular structure data and the molecular property data. Accordingly, the properties of the analyzed molecular data are: IC value of the molecule for the MAT2A target protein 50 value.

[0075] In step S300, the AI ​​agent according to the implementation scheme of the present invention can perform autonomous reasoning based on the user's intention and the nature of the molecular data, using the prior knowledge included in the large language model and the prompt word module to globally generate an operation sequence, that is, to generate an overall idea or conception in a given situation.

[0076] The molecular data processing of drugs can include mining important data in patents or papers, cleaning tabular data containing molecular structures and properties, importing and exporting molecular sets of specific targets from the molecular database of drugs, searching for molecular information of specific structures in the database, converting the representation of one molecule into another, and analyzing the quantitative structure-activity relationship of a specific target compound.

[0077] In drug development, the molecular database of drugs is a database containing information such as molecular structure, various target biological activities and ADMET properties. The molecular data of drugs can use databases such as ChEMBL, PubChem, and REAL. Data from different databases (such as ChEMBL, PubChem, and REAL) may need to be integrated to provide more comprehensive information.

[0078] The ChEMBL database is a manually curated medicinal chemistry database that contains data on drugs and drug-like substances. The ChEMBL database provides detailed target information, efficacy data, and ADMET properties, making it ideal for drug design and repurposing studies. The PubChem database is an open chemical database that contains small molecules, bioactivity data, and related literature. The PubChem database is characterized by its large amount of data and wide coverage, making it suitable for a wide range of scientific research and drug discovery projects. The REAL database is an augmented reality database focused on drugs, providing detailed information on commercialized drugs, including their physicochemical properties, efficacy, and market information.

[0079] After the AI ​​agent understands the user's intention and analyzes the properties of the molecular data, it uses the prior knowledge compressed inside the large language model and the prior knowledge contained in the prompt word module (which contains the task description of various molecular data processing, the calling tool flow of complex tasks, and other prior knowledge) to perform autonomous reasoning to generate a global operation sequence for executing the task, which is a global grasp of the task execution. Therefore, the AI ​​agent according to the present invention can determine the global operation sequence by combining the understanding of the user's intention, the background knowledge of molecular data processing, and the description of relevant facts.

[0080] According to an embodiment of the present invention, the user intends to determine: extract the compounds of MAT2A target in the ChEMBL database and QSAR modeling, and the properties of the analyzed molecular data are: IC values ​​of the molecules for the MAT2A target protein 50 The generated global operation sequence includes: target ID query, natural language-SQL language conversion, molecular data extraction, molecular data cleaning and QSAR modeling.

[0081] In step S400, the purpose of executing a self-consistent loop on the generated global operation sequence is to call the tools in the tool library to process the molecular data of the drug through the sub-steps of thinking, planning, criticizing, acting, and observing. The process of executing a self-consistent loop on the generated global operation sequence is described in detail later.

[0082] After completing a series of analysis and reasoning tasks, the self-consistent loop of the AI ​​agent according to the embodiment of the present invention can output the final processing results in various formats in step S500, which are of great practical value to professionals in the field of drug development and scientific research. The final processing results can include the following forms:

[0083] CSV files containing molecular structures and properties: AI agents output CSV files containing molecular structures and properties through self-consistent loops after data mining, database export, search or analysis. These generated CSV files can help wet lab R&D personnel more intuitively understand the various characteristics of a certain type of molecule with a specific target, structure or property.

[0084] Updated drug molecular database: The structure and property information of specific molecules can be imported into the specified database table by calling the database import tool through the AI ​​agent.

[0085] QSAR model: AI agents can train QSAR models based on information on molecular structure and properties (e.g., biological activity and ADMET properties). The above information can be derived from data in the drug's molecular database, data mining from patents or papers, or input from structure and property data prepared by R&D personnel themselves. The output model can be a QSAR model based on one-dimensional, two-dimensional, or three-dimensional molecular structure.

[0086] Structure of compounds: AI agents can use structure conversion tools to convert different forms of molecular files, such as converting SMILES into CIF, MOL or MOL2, and can also convert images representing molecular structures into the above-mentioned specific molecular structure representations.

[0087] The AI ​​agent according to the embodiment of the present invention can also create interactive data visualization applications. This enables researchers to intuitively explore and analyze data, more effectively understand data patterns and trends, automatically generate reports containing complex data analysis and model results, reduce manual writing time, and improve the accuracy and professionalism of reports.

[0088] In general, the output of AI agents is not limited to a single data or description, but includes multiple forms of information, which together provide users with a comprehensive and in-depth analysis result. Through iterative workflows, AI agents can continuously improve the quality of their output and better meet user needs.

[0089] By combining deep learning technology, natural language processing and professional chemical information processing technology, the AI ​​agent of the present invention provides a powerful tool for drug discovery and molecular biology research. It can not only process and analyze complex data, but also generate intuitive and easy-to-understand results according to needs, greatly accelerating the process of scientific research and drug development.

[0090] Figure 2 It is a flowchart showing that a method for processing molecular data of drugs based on an AI agent according to an embodiment of the present invention performs a self-consistent loop on a generated global operation sequence.

[0091] See also Figure 2 , executing a self-consistent loop on the generated global operation sequence may include: a thinking sub-step S401, a planning sub-step S402, a criticizing sub-step S403, an acting sub-step S404, an observing sub-step S405, and a looping sub-step S406.

[0092] In thinking sub-step S401, the AI ​​agent determines the task type and the execution target of the sub-task according to the global operation sequence.

[0093] The task type can be a drug molecular data processing task that can be completed by a single subtask, or a drug molecular data processing task that requires multiple subtasks to complete. Here, a single subtask can be a task that can be completed using one tool, such as the conversion between molecular data of different drugs; multiple subtasks can be tasks that require multiple tools to complete.

[0094] According to the implementation scheme of the present invention, the generated global operation sequence includes: target ID query, natural language-SQL language conversion, molecular data extraction, molecular data cleaning and QSAR modeling. After thinking about the sub-steps, it can be determined that the task type is multiple sub-tasks, and the goal is to perform target ID query first.

[0095] In the planning sub-step S402, after the AI ​​agent obtains the thinking result in the thinking sub-step S401, it combines the text description of the tool in the tool library module to determine how the sub-action responds (including the selection of the tool), that is, the action sequence of the sub-task. This can include determining the processing object (for example, molecular data of various drugs, including structure, biological activity and ADMET properties), evaluating different action plans, predicting the potential results of different action plans, and selecting the most appropriate single action.

[0096] Different from the global planning of task execution by the operation sequence generated by the AI ​​agent in step S300, the planning sub-step S402 determines the local task planning, i.e., the tool type and calling method of the sub-task, by analyzing the detailed description of each tool in the tool library.

[0097] The planning sub-step S402 can understand the sub-task goal of the target ID query and determine the action sequence to achieve the sub-task, including: using the database target identification tool Target2ID, converting the natural language describing the target into SQL language, and obtaining the database ID of MAT2A after execution.

[0098] The critique sub-step S403 can evaluate the action sequence for implementing the sub-task determined by the plan sub-step S402, and identify missing requirements or contradictions in conjunction with the intermediate state information related to the sub-task stored in the short-term memory component, thereby identifying potential problems, deficiencies or possible improvements, and correcting the action sequence of the sub-task. This helps to provide effective solutions to problems that arise in molecular data processing. For example, for the action sequence of the sub-task "need to use the database target identification tool Target2ID, convert the natural language describing the target into SQL language, and obtain the database ID of MAT2A after execution", potential problems are identified, including: the target text described in natural language is an abbreviation (MAT2A) rather than the full name (ChEMBL ID), which may not be recognized in SQL searches.

[0099] Accordingly, the action sequence of the subtask is modified based on the determined potential problem. For example, the modified action sequence of the subtask may include: converting the natural language describing the target into SQL language using the Text2SQL tool, and performing a query on the target ID using the Target2ID tool.

[0100] In the action sub-step S404, the modified subtask action sequence is executed. Based on the modified subtask action sequence, the natural language describing the target is converted into SQL language using the Text2SQL tool, and the ChEMBL database ID including the MAT2A target is obtained using the Target2ID tool to perform a query on the target ID.

[0101] In the observation sub-step S405, the progress of the sub-task is determined according to the results of the thinking sub-step and the action sub-step, combined with the historical information of the memory module and the prompt word module and the text description of the tool in the tool library module.

[0102] According to an embodiment of the present invention, when the result of thinking is that the molecular data processing task of any drug can be completed by a single subtask, that is, the task can be completed using only one tool, the self-consistent loop ends, and the observation sub-step can output the final processing result of the drug's molecular data.

[0103] When the result of thinking is any drug molecular data processing task that requires multiple subtask data processing to complete, that is, it needs to be completed using multiple tools, it automatically enters the loop sub-step S406 until the multiple subtasks are completed and the final processing result of the drug's molecular data is output.

[0104] In the loop sub-step S406, the new sub-task is continued to be executed based on the result of the thinking sub-step and the result of the observation sub-step. The execution of the new sub-task may include S401 to S405. That is, the execution of each sub-task includes the thinking, planning, criticizing, acting and observing sub-steps. In this way, sub-steps S401 to S406 constitute a self-consistent loop process. This self-consistent loop process reflects that the intelligent agent realizes complex reasoning through intermediate reasoning steps, and collects information through interaction with the external environment to support its decision-making.

[0105] In addition, in order to improve the working effect of the self-consistent cycle of the AI ​​agent, an iterative working mode can be adopted, that is, allowing the AI ​​agent to modify and improve its output multiple times until the task is completed or a certain termination condition is reached or a satisfactory result is achieved. This working mode can significantly improve the performance of AI agents in complex tasks. In general, this self-consistent cycle process enables the agent to learn and adapt through continuous practice, thereby improving its problem-solving ability. In practical applications, this cycle can help the agent better understand and perform complex tasks, especially in situations that require continuous decision-making and adaptation to changing environments.

[0106] The self-consistent loop of the AI ​​agent according to the embodiment of the present invention is a complex system integrating multiple technologies and methods, aiming to improve the ability of the AI ​​agent to handle complex tasks. By combining the deep learning ability of the large language model, the guiding role of the prompt words and the post-processing function of the output parser, it is possible to more effectively understand and perform molecular data processing tasks for various drugs.

[0107] According to the implementation scheme of the present invention, the prompt words, input processing, model reasoning and output parsing can be continuously optimized based on user feedback and evaluation, thereby improving the overall performance of the self-consistent cycle. In the application of AI agents, this synergy greatly improves the learning efficiency of the model and its ability to handle problems in specific fields.

[0108] The following is a specific example to describe the process of implementing drug molecular data processing based on AI agents according to the present invention.

[0109] Figure 3 The processing process of the drug molecular data processing task that can be completed by a single subtask is shown; Figure 4 The SMILES representation of the original drug molecular data is shown; Figure 5It shows the molecular data of the drug after processing by executing a self-consistent cycle using the method for processing the molecular data of the drug based on an AI agent according to the present invention, that is, a single subtask output of converting the SMILES representation of the molecular structure into a molecular structure picture.

[0110] See also Figure 3 The input of the task is the original molecular representation file (mols.csv) and the task description in natural language. The SMILES representation of the molecular structure is as follows: Figure 4 As shown, the natural language description is: I need to convert the SMILES molecular representation in the CSV file just uploaded into a two-dimensional representation of the molecular structure, and save the result in PNG format in / home / user / conversion.

[0111] Specifically, the user's intention is to convert the SMILES molecular representation in the CSV file into a two-dimensional representation of the molecular structure formula, and save the result in PNG format in / home / user / conversion.

[0112] The global operation sequence generated by autonomous reasoning can include: molecular representation conversion. Then enter the self-consistent loop processing. After the thinking sub-step, it can be determined that the task type is a single sub-task, and the goal is to complete the format conversion of the molecular file. After the planning sub-step, it can be determined that the molecular representation conversion tool needs to be used to process the uploaded molecular data of SMILES2Image (the SMILES2Image tool can be implemented using RDKit or other open source APIs), and then the generated image is saved. After the criticism sub-step, it can be determined that there may be a possibility that SMILES represents non-existent or unreasonable molecules, which will cause the task to fail. After the action sub-step, the SMILES2Image tool can be used to process SMILES and save the image in the / home / user / conversion folder. After the observation sub-step, it can be determined that all molecules have been converted, and the image image_den.png containing the two-dimensional representation of the molecular structure is saved in the / home / user / conversion folder. Finally, the final processing result of the molecular data of the drug is presented to the user: the task is completed, and the image image_den.png is saved in the / home / user / conversion folder after being processed by the molecular representation conversion tool.

[0113] The method for processing drug molecular data based on an AI agent according to an embodiment of the present invention utilizes natural language processing technology to parse the user's natural language task description, identify the user's intention and analyze the properties of the molecular data, call internal tools, convert the extracted SMILES code into the corresponding two-dimensional molecular structure, and save it to the user-specified / home / user / conversion folder. The converted image shows the structure of each molecule, providing researchers with intuitive visual information, which is helpful for further chemical analysis and experimental design. The entire process can be completed automatically, reducing the need for manual operation, while improving the accuracy and efficiency of data processing.

[0114] The above steps are only examples, and the actual processing process can be adjusted according to the specific task and the design of the AI ​​agent.

[0115] Figure 6 The figure shows the processing process of the task of processing the molecular data of drugs which can be completed through multiple subtasks according to the method of processing the molecular data of drugs based on AI agents according to an embodiment of the present invention.

[0116] See also Figure 6 , the natural language task description used is: I need to perform QSAR modeling on the compounds of the MAT2A target in the ChEMBL database and save the model in / home / user / qsar. Specifically, the user puts forward a requirement in natural language, requiring QSAR modeling of compounds of a specific target (MAT2A) in the ChEMBL database, and specifies the location where the model is saved.

[0117] The user's natural language task description and the molecular data of the ChEMBL database are input into the constructed AI agent, and the user's intention is determined to be: extract the compound structure and activity information of the MAT2A target in the ChEMBL database, establish a QSAR model based on the above information, and save the model in / home / user / qsar.

[0118] The global operation sequence generated by autonomous reasoning can include: target ID query, natural language-SQL language conversion, molecular data extraction, molecular data cleaning and QSAR modeling.

[0119] After determining the global operation sequence, enter the self-consistent loop processing to process the molecular data of the drug using the tool library module.

[0120] The method for processing molecular data of drugs based on AI intelligent agents according to the implementation scheme of the present invention can determine the action plan through a self-consistent cycle, that is, it is necessary to complete 4 rounds of self-consistent cycles, using database target identification tools, database information extraction tools, data cleaning tools, and QSAR model building tools in sequence to perform the steps of target identification, information extraction, data cleaning, and QSAR model building.

[0121] The database target identification tool Target2ID can accurately identify compound data related to the MAT2A target in the ChEMBL database. The database information extraction tool SQL2CSV can be used to extract detailed information of these compounds from the ChEMBL database, including chemical structure, biological activity, and ADMET properties. The data cleaning tool Normalize can ensure the quality of the extracted data, eliminate any possible errors or inconsistencies, and standardize the data format to meet the needs of QSAR modeling. Finally, the QSAR model building tool CSV2QSAR is used to establish a quantitative structure-activity relationship model based on the cleaned data. This model will be used to predict the biological activity of the compound and assist in drug design and optimization.

[0122] Through the above steps, the molecular data processing method based on AI intelligent drug according to the present invention can complete complex multiple subtasks through multiple rounds of self-consistent cycles.

[0123] Although for the sake of clarity, the exemplary method of the present invention described above is represented as a series of controls, it is not intended to limit the order of execution steps, and multiple steps can be performed simultaneously or in different orders as required. In order to implement the method according to the present invention, the steps shown can further include other steps.

[0124] This process not only demonstrates the practicality of AI agents in the field of chemical information processing, but also reflects its wide application potential in drug design and optimization. Through this automated process, researchers can quickly obtain QSAR models for specific targets, greatly shortening the time required in the early stages of drug development.

[0125] Figure 7 is a block diagram showing a system for processing molecular data of drugs based on an AI agent according to an embodiment of the present invention. Figure 7 According to an embodiment of the present invention, a system for processing molecular data of drugs based on an AI agent may include: a construction module 100, a user intention understanding module 200, an autonomous reasoning module 300, a self-consistent loop module 400 and an output module 500.

[0126] The construction module 100 can be configured to construct an AI agent, which includes a large language model, a prompt word module, a tool library module, a memory module and an output parser. The user intention understanding module 200 can be configured to input the user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data. The autonomous reasoning module 300 can be configured to perform autonomous reasoning based on the user's intention and the properties of the molecular data to generate a global operation sequence. The self-consistent loop module 400 can be configured to perform a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library module. The output module 500 can be configured to output the final processing result of the molecular data of the drug and present it to the user.

[0127] The specific processing process of each module is similar to the method for processing molecular data of drugs based on AI agents according to an embodiment of the present invention, and will not be repeated here.

[0128] In addition, various embodiments of the present invention may be implemented by hardware, firmware, software or a combination thereof. Hardware may be implemented by one or more of an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a general purpose processor, a controller, a microcontroller, a microprocessor, etc.

[0129] The scope of the present invention is also intended to include software or machine executable instructions (e.g., control systems, applications, firmware, programs, etc.) that enable control according to various embodiments to be performed on a device or computer, as well as non-volatile computer readable media executable on a device or computer having such software or instructions, etc. stored thereon.

[0130] The various embodiments of the invention are not an exhaustive list of all possible combinations, but are intended to describe representative aspects of the invention, and what is described in various embodiments may be applied independently or in combinations of two or more.

[0131] The description presented in the above exemplary embodiments is only used to illustrate the technical solution of the present invention, and is not intended to be exhaustive, nor is it intended to limit the present invention to the precise form described. Obviously, it is possible for a person of ordinary skill in the art to make many changes and variations based on the above teachings. The exemplary embodiments are selected and described to explain the specific principles of the present invention and its practical application, so that other technicians in the field can easily understand, implement and use the various exemplary embodiments of the present invention and its various selected forms and modified forms. The scope of protection of the present invention is intended to be limited by the attached claims and their equivalent forms.

Claims

1. A method for processing drug molecular data based on an AI agent, comprising: Constructing an AI agent, wherein the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module, and an output parser; Input the user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data; Based on the user's intent and the nature of the molecular data, autonomous reasoning is performed using the large language model and prior knowledge included in the prompt word module to generate a global action sequence; Execute a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library modules; The final processing results of the drug's molecular data are output and presented to the user.

2. The method for processing drug molecular data based on AI agent according to claim 1, wherein: The large language model is a pre-trained deep learning model, which includes a non-open source large language model and / or an open source large language model, and the large language model is configured to understand the user's natural language and analyze the user's intention; The prompt word module is configured to assist the large language model in understanding and analyzing the user's intention and to determine the molecular data processing task in conjunction with the memory module; The tool library module is configured to integrate tools for processing molecular data of drugs and includes a text description of the tools for processing molecular data of drugs; The memory module is configured to store intermediate state information related to the subtask and historical information of user interactions for enhancing perception, decision-making and action on the molecular data processing task; The output parser is configured to post-process the output generated by the large language model to ensure that the result conforms to an expected format or standard.

3. The method for processing drug molecular data based on AI agent according to claim 2, wherein: Executing a self-consistent loop over a sequence of global operations involves: A thinking sub-step, which determines a task type and an execution target of a sub-task according to a global operation sequence, wherein the task type includes a drug molecular data processing task completed by a single sub-task and a drug molecular data processing task completed by multiple sub-tasks; The planning sub-step combines the text description of the tool in the tool library module to determine the action sequence of the sub-task; The critique sub-step evaluates the action sequence of the subtask, combines the information stored in the memory module to determine potential problems, and corrects the action sequence of the subtask; an action sub-step, which utilizes the tool library module to execute the modified action sequence of the sub-task; The observation sub-step determines the progress of the task based on the results of the thinking sub-step and the action sub-step, combined with the historical information of the memory module and the prompt word module and the text description of the tool in the tool library module.

4. The method for processing drug molecular data based on AI agent according to claim 3, wherein: Executing a self-consistent loop over the sequence of global operations further comprises: When the task type determined by the thinking sub-step is a task of processing the molecular data of a drug completed by a single sub-task, the observing sub-step outputs the final processing result of the molecular data of the drug; When the task type determined in the thinking sub-step is a task of processing the molecular data of a drug to be completed by multiple sub-tasks, the loop sub-step is automatically entered until the multiple sub-tasks are completed and the final processing result of the molecular data of the drug is output; The loop sub-step is configured to continue executing a new sub-task based on the result of the thinking sub-step and the result of the observing sub-step, wherein executing the new sub-task includes a thinking sub-step, a planning sub-step, a criticizing sub-step, an acting sub-step and an observing sub-step.

5. The method for processing drug molecular data based on AI agent according to claim 2, wherein: The prompt word module is configured as follows: Includes a description of the tasks involved in processing molecular data on drugs; Includes task descriptions for processing drug molecular data and prior knowledge of tool flows for complex tasks; Leverage hierarchical cues to guide large language models to learn step-by-step.

6. The method for processing drug molecular data based on AI agent according to claim 2, wherein: The tool library module includes an application programming interface for the underlying algorithm; The tool library module further includes at least one of a data mining tool, a data cleaning tool, a database import and export tool, a data search tool, a data conversion tool and a data analysis tool for processing the molecular data of the drug.

7. The method for processing drug molecular data based on AI agent according to claim 3, wherein: The memory module includes a long-term memory component and a short-term memory component; The long-term memory component is configured to store data related to the user interaction history; The short-term memory component is configured to store intermediate state information associated with the subtask for use in identifying potential problems in the critique substep of the self-consistent loop.

8. The method for processing drug molecular data based on AI agent according to claim 1, wherein: The user's natural language task description includes at least one of operations for extracting, storing, mining, predicting, visualizing, and searching for molecular data of drugs; The molecular data includes molecular structure data and files containing molecular structure data and molecular property data; The molecular structure data includes at least one of a one-dimensional representation, a two-dimensional representation, and a three-dimensional representation of the molecular structure, and the molecular property data includes at least one of the properties of activity, absorption, distribution, metabolism, excretion, and toxicity of the drug; Among them, the one-dimensional representation of molecular structure includes files in SMILES format; the two-dimensional representation of molecular structure includes molecular pictures in JPG and PNG or molecular files of atoms and chemical bond positions in MOL and SDF; the three-dimensional representation of molecular data includes molecular files in PDB, CIF and MOL2; files containing molecular structure data and molecular property data include list files and database files, list files include CSV and XLSX files, and database files include SQL files.

9. The method for processing drug molecular data based on AI agent according to claim 8, wherein: The final processing results of the AI ​​agent include: at least one of a one-dimensional representation, a two-dimensional representation and a three-dimensional representation of the molecular structure, a file containing molecular structure data and molecular property data, and a QSAR model of the drug molecule.

10. A system for processing molecular data of drugs based on AI agents, comprising: A construction module configured to construct an AI agent, wherein the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module, and an output parser; A user intention understanding module configured to input a user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data; an autonomous reasoning module configured to perform autonomous reasoning based on user intention and properties of the molecular data using the large language model and prior knowledge included in the prompt word module to generate a global operation sequence; a self-consistent loop module configured to execute a self-consistent loop on a global operation sequence to process molecular data of a drug using a tool library module; The output module is configured to output the final processing result of the molecular data of the drug and present it to the user.

11. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the following steps are implemented: Constructing an AI agent, wherein the AI ​​agent includes a large language model, a prompt word module, a tool library module, a memory module, and an output parser; Input the user's natural language task description and the molecular data of the drug to be processed into the constructed AI agent to understand the user's intention and analyze the properties of the molecular data; Based on the user's intent and the nature of the molecular data, autonomous reasoning is performed using the large language model and prior knowledge included in the prompt word module to generate a global action sequence; Execute a self-consistent loop on the global operation sequence to process the molecular data of the drug using the tool library modules; The final processing results of the drug's molecular data are output and presented to the user.

Images