A large model-driven embodied intelligent biosynthesis system and method

Abstract

The present invention discloses a large-model driven embodied intelligent biosynthesis system and method. The system adopts a centralized multi-large-model intelligent agent architecture and performs tasks through modular intelligent agents. The system consists of a human-computer collaborative interaction module, a global task management module, a general information retrieval module, a reverse path planning module, a protease modification module, a metabolic pathway optimization module, a wet experiment embodied execution module and a code execution module. Each module integrates different tools and data sources for a specific task field. The present invention can autonomously complete complex biosynthesis experiment design, planning and execution, and can dynamically adapt to functional modules and wet experiment feedback to ensure the flexibility of experimental planning.

Description

Technical Field

[0001] The present invention relates to the field of autonomous biosynthesis technology, and more specifically to a large model-driven embodied intelligent biosynthesis system and method. Background Art

[0002] Biosynthesis gradually builds chemical structures through enzymatic reactions and is one of the core and foundations of molecular creation. Compared with traditional chemical synthesis, biosynthesis has unique advantages such as high catalytic activity, strong specificity, and mild reaction conditions, and plays a vital role in promoting the revitalization of the bioeconomy. However, current biosynthesis experiments mainly rely on the paradigm of artificial trial and error: experimenters computer-aided or rationally design synthesis pathways, use gene editing or metabolic engineering to construct chassis cells or enzymatic reaction systems, and then verify the efficiency and accuracy of product generation through wet experiments. The process is cumbersome to operate, the experimental cost is high, and the efficiency and success rate of the design are low, which greatly limits the widespread application of biosynthesis technology.

[0003] Embodied intelligence is the intersection of machine intelligence and embodied cognition. It believes that intelligent behavior is not generated by the brain (software) alone, but is facilitated by the coordination between the brain (software), the body (hardware), and the environment. Embodied intelligence systems interact with the environment from a "first-person" perspective, perceive and manipulate the environment in real time through their own physical presence (such as robotic arms and sensors), and complete preset tasks autonomously. Embodied intelligence systems have demonstrated a high degree of autonomy and environmental adaptability in tasks such as object manipulation and vision-language navigation (VLN), triggering changes in research and application in fields such as computer vision, natural language processing, and robotics.

[0004] In recent years, laboratory automation technology has gradually been applied to the field of biosynthesis. Automated experimental platforms can perform standardized liquid handling tasks such as pipetting, mixing, thermal cycling, etc. through programming interfaces. Such equipment significantly improves experimental efficiency and reduces errors introduced by manual operations. However, although existing platforms can automatically perform preset tasks, they still have obvious deficiencies in embodied intelligence: 1) Lack of closed-loop optimization capabilities: existing systems cannot dynamically adjust experimental processes based on experimental feedback; 2) Lack of the ability to autonomously plan experimental steps: The operating procedures of existing systems still require manual pre-writing of instructions and codes, and the degree of autonomy is limited; 3) Lack of the ability to autonomously design synthetic pathways and enzyme sequences: The reaction system composed of substrates and enzymes relies on artificial rational design or computer-aided design, and automated planning has not yet been achieved.

[0005] Therefore, how to provide an embodied intelligent biosynthesis system that can autonomously plan biosynthesis experimental plans, perform wet experimental operations, and dynamically adjust plans based on execution feedback is an urgent problem to be solved by people in this field. Summary of the invention

[0006] In view of this, the present invention provides a large model-driven embodied intelligent biosynthesis system and method, which can solve the technical problems described in the background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, an embodiment of the present invention provides an embodied intelligent biosynthesis system driven by a large model, the system adopts a centralized multi-large model intelligent agent architecture, and the system includes: a human-computer collaborative interaction module, a global task management module, a reverse path planning module, a protease modification module, a metabolic pathway optimization module, a general information retrieval module, a wet experiment embodied execution module and a code execution module; wherein:

[0009] The human-computer collaborative interaction module is used to receive the experimental requirements input by the user and transmit the experimental requirements to the global task management module;

[0010] The global task management module is driven by a large model and is used to decompose experimental requirements into multiple tasks and schedule them to be executed by various functional modules;

[0011] The reverse pathway planning module is used to infer the reaction precursors and search for substrate molecules based on the molecules to be synthesized input by the user, so as to generate the biosynthetic reaction pathway of the target product;

[0012] The protease modification module selects the natural enzyme amino acid sequence based on the biosynthetic reaction pathway generated by the reverse pathway planning module, and generates a corresponding enzyme amino acid skeleton sequence modification scheme in combination with the synthesis target input by the user;

[0013] The metabolic pathway optimization module optimizes the metabolic pathway in an in vitro or in vivo environment based on the generated biosynthetic reaction pathway and enzyme amino acid sequence;

[0014] The general information retrieval module uses a large model-driven search engine to call external data sources to obtain biosynthesis knowledge and assist in generating wet experiment operation steps and plans;

[0015] The wet experiment embodied execution module is used to convert the output wet experiment scheme into the control code of the embodied biosynthesis device, execute the wet experiment, and dynamically observe the experimental process and experimental results;

[0016] The code execution module is used to call the code compiler interface to perform calculation tasks of simulating experimental conditions, analyzing experimental data, and optimizing experimental parameters;

[0017] The human-computer collaborative interaction module, reverse path planning module, protease modification module, metabolic pathway optimization module, general information retrieval module, wet experiment embodied execution module and code execution module are all started or ended by the global task management module. After completion, the operating status and observations are fed back to the global task management module. The global task management module dynamically adjusts and optimizes the experimental plan according to the feedback information of each functional module, completes the overall planning and management of the task, and realizes the closed-loop optimization of dry and wet experiments in autonomous biosynthesis.

[0018] In a preferred embodiment, the human-computer collaborative interaction module is also used to visualize the execution progress of the solution, the task status and any changes according to the task planning and execution description file of the global task management module.

[0019] In a preferred embodiment, the global task management module is developed based on the Python Flask or Django backend framework and integrates a large model API; it adopts a combination of ReAct prompts and zero-sample thinking chain reasoning methods to dynamically call each functional module during the reasoning process; the wet experiment requirement instruction information sent by the human-computer collaborative interaction module and from the user is decomposed into multiple executable subtasks. During the task execution process, feedback information from other functional modules is received through the REST API interface, and task planning and decomposition plans are dynamically adjusted accordingly.

[0020] In a preferred embodiment, the reverse pathway planning module infers and designs the optimal metabolic pathway from the starting material to the target product through large model-driven enzymatic reaction database retrieval and reverse pathway planning tools.

[0021] In a preferred embodiment, the protease modification module screens and optimizes key enzymes in biosynthesis based on protein language models, structure prediction tools, and enzyme kinetics simulation tools.

[0022] In a preferred embodiment, the metabolic pathway optimization module optimizes and transforms the metabolic network of the chassis in an in vivo environment in combination with the genome-scale metabolic network according to the inherent metabolic environment of the microbial chassis.

[0023] In a preferred embodiment, the general information retrieval module integrates a large model API, an external search engine, and an internal vector database to retrieve experimental data and literature that are highly relevant to the current biosynthesis task based on semantic similarity.

[0024] In a preferred embodiment, the wet experiment embodied execution module integrates a biosynthetic embodied experiment platform, a control document vector library, and a large model API, wherein:

[0025] The biosynthesis embodied experimental platform is used to separate the preparation of liquid systems from the analysis of liquid systems. The preparation of liquid systems uses an open source programmable liquid robot platform equipped with a pipette, a gripper, a heating and shaking module, a temperature control module, a magnetic control module, and a thermal cycle module; the analysis of liquid systems uses an ELISA reader, a spectrometer, and a chromatograph to monitor the experimental process in real time.

[0026] The control document vector library stores the vector representation of the low-level and high-level control code development documents of each experimental device encoded by the multimodal large model;

[0027] The large model combines input instructions with the retrieval-enhanced generation method to automatically generate control codes for the experimental platform and perform biological synthesis wet experiments; the synthesis and analysis results of the experiment are fed back to the global task management module through the REST API.

[0028] In a preferred embodiment, the code execution module receives task codes from other functional modules through the REST API interface, and the task codes are passed into the Docker container for execution. After execution, the results are fed back to the global task management module through the REST API.

[0029] In a second aspect, an embodiment of the present invention further provides a large model-driven embodied intelligent biosynthesis method, using the above large model-driven embodied intelligent biosynthesis system to achieve autonomous biosynthesis, the method comprising:

[0030] Receive the experimental requirements input by the user through the human-computer collaborative interaction module, and transmit the experimental requirements to the global task management module;

[0031] Use the global task management module to decompose the experimental requirements into multiple tasks, dispatch them to each functional module for execution, and dynamically adjust and optimize the experimental plan based on the feedback information of each functional module to complete the overall planning and management of the task;

[0032] The reverse pathway planning module is used to infer the reaction precursors and search for substrate molecules based on the molecules to be synthesized input by the user, thus generating the biosynthetic reaction pathway of the target product;

[0033] The protease modification module selects the natural enzyme amino acid sequence based on the biosynthetic reaction pathway and generates the corresponding enzyme amino acid skeleton sequence modification scheme in combination with the synthesis target input by the user;

[0034] The metabolic pathway optimization module optimizes metabolic pathways in vitro or in vivo based on the generated biosynthetic reaction pathways and enzyme amino acid sequences;

[0035] Through the general information retrieval module, a large model-driven search engine is used to call external data sources to obtain biosynthetic knowledge and assist in generating wet experiment operation steps and plans;

[0036] The wet experiment embodied execution module is used to convert the output wet experiment plan into the control code of the embodied biosynthesis equipment, execute the wet experiment, and dynamically observe the experimental process and experimental results;

[0037] The code execution module is used to call the code compiler interface to perform computational tasks such as simulating experimental conditions, analyzing experimental data, and optimizing experimental parameters.

[0038] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following technical advantages:

[0039] 1. The present invention provides an embodied intelligent biosynthesis system driven by a large model, which can autonomously plan biosynthesis experimental plans, perform wet experimental operations, and dynamically adjust plans based on execution feedback, which is conducive to ensuring the flexibility of experimental planning and helping to improve the degree of autonomy of biosynthesis.

[0040] 2. Closed-loop optimization: The present invention adopts a centralized multi-model intelligent agent architecture, and the global task management module adopts ReAct prompt technology and zero-sample thinking chain technology to realize dynamic scheduling of each functional module. The global task management module can adjust the experimental process according to the feedback of each functional module, and the module feedbacks the operating status and experimental observation results through the API interface. The system allows a certain flexibility in task execution, does not limit the strict execution order, ensures closed-loop optimization of the experimental process during the experiment, and realizes closed-loop control of the entire process of autonomous biosynthesis.

[0041] 3. Ability to autonomously plan experimental steps: The present invention receives the experimental requirements input by the user through the human-computer collaborative interaction module, and submits it to the global task management module for task decomposition. The global task management module decomposes the experimental task into multiple subtasks and dispatches them to each functional module for execution to ensure the autonomous generation of experimental steps. In particular, the wet experiment embodied execution module stores the vector representation of the underlying and high-level control code development documents of the experimental equipment through the integrated control document vector library and large model API. The system automatically generates the control code of the experimental platform through the retrieval enhanced generation (RAG) technology, combined with the experimental requirements and steps input by the user, to achieve autonomous execution of experimental operations without human intervention.

[0042] 4. Ability to autonomously design synthesis paths and enzyme sequences: The present invention can automatically generate synthesis paths and enzyme sequence design schemes through the scheduling capability of the global task management module, combined with the reverse synthesis path planning tool and the large model API. In particular, the reverse path planning module integrates the reverse synthesis tool and the general information retrieval module, which can automatically infer the reaction precursors and substrate molecules according to the user input and generate the biosynthetic reaction path of the target product. At the same time, the protease modification module integrates the enzyme sequence vector database, enzyme selection tool, structure prediction tool and molecular dynamics simulation tool, combined with the large model API to generate the corresponding enzyme sequence optimization and modification scheme, thereby realizing automated enzyme design and optimization.

[0043] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0044] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0047] Figure 1 A schematic diagram of the architecture of a large model-driven embodied intelligent biosynthesis system provided in an embodiment of the present invention.

[0048] Figure 2 A schematic diagram of a Json file describing task planning and execution of a global task management module provided in an embodiment of the present invention.

[0049] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] In the description of the present invention, it should be noted that: in some processes described in the specification and drawings of this application, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be performed in the order in which they appear in this document or may be performed in parallel. In addition, various serial numbers are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0052] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0053] See also Figure 1 As shown in the figure, considering that although the field of biosynthesis has made some progress in experimental automation, there are still obvious deficiencies in embodied intelligence. The present invention proposes an embodied intelligent biosynthesis system driven by large language models (LLMs) to solve the problem of insufficient autonomy in biosynthesis. The system consists of eight modules: human-computer collaborative interaction module, global task management module, general information retrieval module, reverse path planning module, protease modification module, metabolic pathway optimization module, wet experiment embodied execution module, and code execution module. It can independently complete complex experimental design, planning and execution, and can dynamically adapt to functional modules and wet experiment feedback to ensure the flexibility of experimental planning.

[0054] The system of the present invention is based on a centralized multi-large model agent architecture (Multi-LLM-Agents), and performs tasks through modular agents. Each module integrates different tools and data sources for specific task areas. The core global task management module is responsible for the scheduling and coordination between agents to ensure the reasonable allocation of system resources and the efficient execution of tasks. The following is a detailed introduction to the eight main modules of the system of the present invention:

[0055] Human-computer collaborative interaction module: This module provides an interface for users to interact with the system. Users can input experimental design requirements through this module. The system will generate a preliminary experimental plan through a large model and continuously optimize the experimental process based on real-time feedback from users. The human-computer collaborative interaction module ensures the flexibility and accuracy of experimental planning, allowing users to dynamically adjust the experimental design to ensure the accurate realization of experimental requirements.

[0056] Global task management module: This module is the core scheduling unit of the entire system and is driven by a large language model (such as GPT-4o). The global task management module is responsible for processing user input and feedback information from each submodule, and for refining, decomposing, and scheduling resources for biosynthesis tasks. This module completes the overall planning and management of tasks by dynamically calling each submodule, and continuously optimizes the design process based on experimental feedback.

[0057] General information retrieval module: A search engine driven by a large model (such as PubMed command) calls external data sources (including scientific research literature, patents, and synthetic experimental documents), combined with retrieval enhancement generation (RAG) technology to provide decision support for other modules in the system. This module provides comprehensive data support for biosynthesis tasks through integrated retrieval of multi-source data.

[0058] Reverse Pathway Planning Module: This module uses large model-driven enzymatic reaction database retrieval and reverse path planning tools to infer and design the optimal metabolic pathway from the starting material to the target product. This module provides accurate metabolic pathway planning based on existing biological data for experimental design, ensuring the feasibility and efficiency of pathway design.

[0059] Protease modification module: The protease optimization module relies on protein language models, structure prediction tools, and enzyme kinetics simulation tools to screen and optimize key enzymes in biosynthesis. Through this module, the system can optimize enzyme selection and reaction conditions for different biosynthesis tasks to ensure the efficiency and success rate of biological reactions.

[0060] Metabolic pathway optimization module: This module optimizes metabolic pathways in vitro and in vivo through large model reasoning capabilities and combines multiple data sources (such as metabolic network data and genomic information). In the in vitro environment, the conditions are highly controllable, and the optimal enzymatic reaction conditions can be inferred and achieved; in the in vivo environment, combined with the genome-scale metabolic network (GEM), the system can optimize and transform the metabolic network of the chassis according to the inherent metabolic environment of the microbial chassis to adapt to specific biosynthetic tasks.

[0061] Wet experiment embodied execution module: This module generates control codes (including low-level and high-level controls) for embodied experimental hardware based on the designed synthesis path, enzyme selection and metabolic pathway optimization results, and autonomously executes the experimental plan. Through actual wet experiments, the experimental results such as biosynthetic product yield and reaction efficiency are verified, and the results are fed back to the system to further optimize the experimental design.

[0062] Code execution module: This module is responsible for the scientific computing and data processing tasks involved in the experimental design process. By calling the code compiler interface (such as PYTHON commands), this module can perform computationally intensive tasks such as simulating experimental conditions, analyzing experimental data, and optimizing experimental parameters, providing computational support for the design and adjustment of the entire synthesis plan.

[0063] The working principle of each module of the present invention and its construction method are described in detail below:

[0064] like Figure 1 As shown in the figure, the system adopts a centralized multi-model intelligent agent architecture, receives the experimental requirements input by the user through the human-computer collaborative interaction module, decomposes them into multiple tasks by the global task management module, and dispatches them to various functional modules for execution; the reverse path planning module infers the reaction precursors and searches for substrate molecules based on the molecules to be synthesized input by the user, and generates the biosynthetic reaction pathway of the target product; the protease modification module selects the appropriate natural enzyme amino acid sequence based on the pathway generated by the reverse path planning module, and generates the corresponding enzyme amino acid skeleton sequence modification plan based on the synthesis target input by the user; the metabolic pathway optimization module optimizes the metabolic pathway in vitro or in vivo based on the generated biosynthetic pathway and enzyme amino acid sequence; the general information retrieval module is responsible for acquiring biosynthetic knowledge from external data sources and assisting in generating wet experiment operation steps and plans; the wet experiment embodied execution module converts the upstream output of the wet experiment into The scheme is converted into the control code of the embodied biosynthesis equipment, the wet experiment is executed, and the experimental process and results are dynamically observed; the code execution module is responsible for the scientific calculation and data processing tasks involved in the experimental design process; the above-mentioned functional modules are all scheduled to start or end by the global task management module, and after the end, the running status is fed back and observed to the global task management module through the API interface; although there is a certain sequence dependency between the functional modules, the system does not limit the strict task execution order, so that the global task management module can dynamically adjust and optimize the experimental scheme according to the feedback of each module, and finally realize the closed-loop optimization of dry and wet experiments of autonomous biosynthesis; the human-computer collaborative interaction module visualizes the scheme execution progress, task status and any changes of the embodied intelligent biosynthesis system according to the task planning and execution description file of the global task management module, to ensure real-time monitoring by users and to intervene in the experimental process when necessary.

[0065] In a specific embodiment, the construction process of the embodied intelligent biosynthesis system driven by a large model includes: constructing a human-computer collaborative interaction module; constructing a global task management module; constructing a general information retrieval module; constructing a reverse path planning module; constructing a protease modification module; constructing a metabolic pathway optimization module; constructing a wet experiment embodied execution module; and constructing a code execution module. Among them;

[0066] Preferably, a human-computer collaborative interaction module is constructed, including: the module is constructed using a Web front-end framework (such as React or Vue.js), providing a text input interface, allowing users to input biosynthesis experiment requirements, such as molecules to be synthesized, desired properties, whether to use a microbial chassis, and the type of microbial chassis; the user's text input is transmitted to the global task management module through the REST API interface for subsequent text parsing, task planning and decomposition; the module receives the .json format task planning and execution description file from the global task management module in real time through the WebSocket protocol, and uses dynamic data binding technology (such as Vuex or Redux) to display the task execution progress, task status and any changes of the embodied intelligent biosynthesis system in real time; the user can intuitively monitor the various experimental operations of the system, and when necessary, enter new instructions or adjust parameters through the interface to achieve intervention and control of the experimental process.

[0067] Preferably, a global task management module is constructed, including: the module adopts ReAct prompt technology, and by combining the Zero-shot Chain-of-Thought (Zero-Shot-CoT) reasoning method, allows dynamic calling of each functional module during the reasoning process, and decomposes the wet experiment requirement instruction information from the user sent by the human-computer collaborative interaction module into multiple executable subtasks. During the task execution process, the module receives feedback information from other functional modules through the REST API interface, and dynamically adjusts the task planning and decomposition plan accordingly; the module is developed based on the Python Flask or Django backend framework, and integrates a large model API (such as GPT-4o) to enhance reasoning and decision-making capabilities; the planning and decomposition of tasks are managed through the task planning and execution description file in .json format, such as Figure 2 As shown in the figure, the file records in detail the execution steps, dependencies, called tools and feedback mechanisms of the experiment; the decomposition of tasks and the calling of tools are scheduled through the task queue system (such as Celery or Redis Queue), and the task status and feedback are managed through the message queue system (such as RabbitMQ or Kafka) to ensure the orderly execution of tasks and real-time feedback.

[0068] Preferably, the general information retrieval module includes: the module adopts retrieval-augmented generation (RAG) technology, integrates large model API (such as GPT-4o), external search engine and internal vector database; the external search engine API includes academic search engines (such as Google Scholar, PubMed, Google Patents, USPTO) and general search engines (such as Google, Baidu); the internal vector database stores vector representations of scientific research materials and experimental documents encoded by multimodal large models (such as CLIP), including text data such as scientific research papers, technical patents, experimental reports, scientific research books, as well as image and document data such as experimental logs, synthesis plans, chassis configurations, synthesis results, etc.; the module quickly retrieves experimental data and literature that are highly relevant to the current biosynthesis task based on semantic similarity, thereby expanding the information source of each biosynthesis functional module, enhancing the reasoning and decision-making ability of the module, and assisting in the generation of wet experiment operation steps and plans.

[0069] Preferably, a reverse path planning module is constructed, including: the module adopts ReAct prompt technology, integrates rule-based reverse synthesis path planning tools (such as RetroPath2.0), neural network-based reverse synthesis path planning tools (such as BioNavi-NP, ECREACT), large model API (such as GPT-4o) and general information retrieval module; the module receives reverse path planning instructions from the global task management module through the REST API, and reasoning under the enhancement of the general information retrieval module, selects tools to generate possible biosynthesis paths, and feeds the paths back to the global task management module.

[0070] Preferably, a protease modification module is constructed, including: the module adopts ReAct prompt technology to receive the planned biosynthetic pathway and enzyme selection / modification instructions from the global task management module; the module integrates an enzyme sequence vector database, a biosynthetic pathway-guided enzyme selection tool (such as Selenzyme), a structure prediction tool (such as AlphaFold), a molecular dynamics simulation tool, and a large model API (such as GPT-4o); the enzyme selection tool screens out potential enzymes that meet the target reaction from the existing enzyme library; the enzyme sequence vector database stores the vector representation of enzyme sequences (such as UniProt) encoded by a protein large model (such as ESM-2), and the module expands the potential enzyme space based on homology similarity retrieval; the structure prediction tool (such as AlphaFold) generates a three-dimensional structural model of the enzyme and preliminarily evaluates the structural stability; the molecular dynamics simulation tool simulates the dynamic behavior of the enzyme under specific reaction conditions to derive possible modification schemes; the amino acid sequence of the enzyme modified by the module is fed back to the global task management module through the REST API.

[0071] Preferably, a metabolic pathway optimization module is constructed, including: the module adopts ReAct prompt technology to receive planned biosynthetic pathway optimization instructions including enzyme sequences from the global task management module; the module integration includes enzyme kinetics data sets, machine learning tools, genome-scale metabolic network (GEM) database, metabolic network analysis tools and large model API (such as GPT-4o); machine learning tools learn the mapping relationship between enzymatic reactions and reaction conditions based on enzyme kinetics data sets (such as BKMS-react); metabolic network analysis tools (such as metabolic balance analysis, FBA) are based on the high-quality genome-scale metabolic network database BiGG, and optimize the chassis metabolic network under the guidance of preset biosynthetic goals (such as maximizing the chassis biosynthesis yield under the condition of maintaining normal chassis growth); the module metabolic pathway optimization results are fed back to the global task management module through the REST API.

[0072] Preferably, a wet experiment embodied execution module is constructed, including: the module adopts ReAct prompt technology to receive wet experiment schemes and experimental steps from the global task management module; the module integrates a control document vector library, a biosynthesis embodied experiment platform and a large model API (such as GPT-4o); the biosynthesis embodied experiment platform separates liquid system preparation from liquid system analysis, and the liquid system preparation uses an open source programmable liquid robot platform (such as Opentrons-Flex), equipped with liquid handling modules such as pipettes, grippers, heating and shaking modules, temperature control modules, magnetic control modules, and thermal cycle modules; liquid system analysis uses analysis modules such as microplate readers, spectrometers, and chromatographs. Through an open source programmable robotic arm, the platform realizes flexible movement and operation of liquids between system preparation and system analysis, and is equipped with sensing devices such as visible light cameras, infrared cameras, and temperature sensors to monitor the experimental process in real time; the control document vector library stores vector representations of the underlying and high-level control code development documents of each experimental device encoded by a multimodal large model (such as CLIP); the large model combines input instructions and uses retrieval-augmented generation (Retrieval-Augmented Generation). The Biosynthesis and Analysis (Biosynthesis) technology is used to automatically generate control codes for the experimental platform and perform biological synthesis wet experiments. The synthesis and analysis results of the experiments are fed back to the global task management module through the REST API to achieve dynamic optimization of the experiments and result feedback.

[0073] Preferably, a code execution module is constructed, including: the module uses the Docker API to implement isolated programming environment support to protect the user host from possible unexpected operations; the module pre-builds a programming environment configuration (such as Python, R, MATLAB) that supports scientific computing and data processing tasks; the module receives programming code from other functional modules through the REST API interface, and the task code is passed into the Docker container for execution. After execution is completed, the result is fed back to the global task management module through the REST API.

[0074] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a large-model driven embodied intelligent biosynthesis system: the system adopts a centralized multi-large-model intelligent agent architecture, and decomposes the user instructions from the human-computer collaborative interaction module into multiple tasks based on the global task management module, and dispatches them to each functional module for execution; the core global task management module combines ReAct prompt technology and zero-sample thinking chain technology, does not limit the strict execution order between functional modules, and can flexibly adjust the experimental plan according to the execution feedback of the functional modules; most functional modules enhance the decision-making level of large-model intelligent agents in the field of biosynthesis by integrating special tools, databases or wet experiment embodied hardware equipment in the field of biosynthesis; the human-computer collaborative interaction module provides a visual interface, allowing human scientists to intuitively monitor the experimental process, and when necessary, input new instructions or adjust parameters through the interface to achieve intervention and control of the experimental process. The main working scenarios of the present invention are designing biosynthetic pathways of target molecules, designing cell-free (in-vitro) biosynthetic schemes of target molecules, and other biomanufacturing scenario requirements. It can independently complete complex experimental design, planning, and execution, dynamically adapt to functional modules and wet experiment feedback, improve the degree of autonomy of biosynthesis, lower the threshold for biosynthetic research, and promote the widespread application of biosynthetic technology in the process of bioeconomic revitalization.

[0075] Furthermore, an embodiment of the present invention also provides a large model-driven embodied intelligent biosynthesis method, which uses a large model-driven embodied intelligent biosynthesis system in the above embodiment to achieve autonomous biosynthesis, and the method mainly includes:

[0076] Receive the experimental requirements input by the user through the human-computer collaborative interaction module, and transmit the experimental requirements to the global task management module;

[0077] Use the global task management module to decompose the experimental requirements into multiple tasks, dispatch them to each functional module for execution, and dynamically adjust and optimize the experimental plan based on the feedback information of each functional module to complete the overall planning and management of the task;

[0078] The reverse pathway planning module is used to infer the reaction precursors and search for substrate molecules based on the molecules to be synthesized input by the user, thus generating the biosynthetic reaction pathway of the target product;

[0079] The protease modification module selects the natural enzyme amino acid sequence based on the biosynthetic reaction pathway and generates the corresponding enzyme amino acid skeleton sequence modification scheme in combination with the synthesis target input by the user;

[0080] The metabolic pathway optimization module optimizes metabolic pathways in vitro or in vivo based on the generated biosynthetic reaction pathways and enzyme amino acid sequences;

[0081] Through the general information retrieval module, a large model-driven search engine is used to call external data sources to obtain biosynthetic knowledge and assist in generating wet experiment operation steps and plans;

[0082] The wet experiment embodied execution module is used to convert the output wet experiment plan into the control code of the embodied biosynthesis equipment, execute the wet experiment, and dynamically observe the experimental process and experimental results;

[0083] The code execution module is used to call the code compiler interface to perform computational tasks such as simulating experimental conditions, analyzing experimental data, and optimizing experimental parameters.

[0084] The specific application of the system of the present invention is described below by taking the cell-free (in-vitro) biosynthesis of "L-lactic acid" as an example:

[0085] L-lactic acid is widely used in food, medicine, biomaterials and other fields. Traditional L-lactic acid synthesis mainly relies on microbial fermentation, but due to the complexity of the microbial metabolic network and the limitations of the cellular environment, the reaction conditions are difficult to flexibly control, which affects the yield and efficiency. Cell-free biosynthesis can bypass the constraints of the microbial chassis, provide greater flexibility, and optimize product synthesis by precisely controlling reaction conditions (such as temperature and substrate concentration). However, traditional cell-free synthesis processes usually rely on fixed reaction conditions and are difficult to optimize and adjust based on experimental feedback.

[0086] The present invention proposes a large model-driven embodied intelligent biosynthesis system that can dynamically adjust reaction conditions according to experimental feedback to achieve automation and optimization of the L-lactic acid synthesis process. The following is the specific process of this embodiment:

[0087] S1. User command input:

[0088] The user inputs the command "Please biosynthesize L-lactic acid in a cell-free system" through the human-computer collaborative interaction module. The command is transmitted to the global task management module through the REST API interface.

[0089] S2. Synthesis task planning and decomposition:

[0090] The global task management module receives user instructions and calls the large model interface to decompose the task into multiple subtasks, which are reverse path planning, protease modification, metabolic pathway optimization, general information retrieval, and wet experiment embodiment execution. The planning and execution description of the task is stored in JSON format, and the reverse path planning module is scheduled first;

[0091] S3. Reverse biosynthesis pathway planning:

[0092] S31, the reverse path planning module receives the instruction "planning the biosynthesis path of L-lactic acid" from the global task management module, and calls the large model interface for reasoning;

[0093] S32. The large model of the reverse path planning module calls the RetroBioCat tool to determine that lactate dehydrogenase (LDH) catalyzes pyruvate to produce L-lactic acid. The reaction equation is as follows:

[0094] Pyruvate + NADH + H + →L-lactic acid+NAD + ;

[0095] S32, the reverse path planning module returns the generated path to the global task management module through the API, and the global task management module determines that the reverse path planning module has completed the task, updates the task planning and execution description file, and schedules the protease transformation module;

[0096] S4. Protease selection:

[0097] S41, the protease modification module receives the instruction from the global task management module: "The biosynthetic pathway of L-lactic acid is known: pyruvate + NADH + H + →L-lactic acid+NAD + , requiring the selection of a reasonable protease for the pathway”, calling the large model interface reasoning;

[0098] S42, the large model of the protease modification module calls the Selenzyme tool to select the LDH enzyme that is most suitable for the catalytic reaction, and then calls AlphaFold to generate the three-dimensional structure of LDH and perform a preliminary structural stability assessment;

[0099] S43, the protease transformation module returns the generated path to the global task management module through the API, and the global task management module determines that the protease transformation module has completed the task, updates the task planning and execution description file, and schedules the metabolic pathway optimization module;

[0100] S5. Metabolic pathway optimization:

[0101] S51, the metabolic pathway optimization module receives the instruction from the global task management module: "The biosynthetic pathway of L-lactic acid is known to be: pyruvate + NADH + H + →L-lactic acid+NAD + , and the corresponding enzyme sequence and structure, requiring the optimization of metabolic pathways in a cell-free reaction system", calling the large model interface reasoning;

[0102] S52. The large model of the metabolic pathway optimization module uses enzyme kinetic data and machine learning tools to derive the optimal reaction conditions: the buffer solution has a pH value of 6.5, the reaction system temperature is 25°C, the initial concentration of pyruvate is 10mM, and the initial concentration of NADH is 1mM;

[0103] S53, the metabolic pathway optimization module returns the generated path to the global task management module through the API, and the global task management module determines that the metabolic pathway optimization and transformation module has completed the task, updates the task planning and execution description file, and schedules the general information retrieval module;

[0104] S6. Wet experiment plan generation:

[0105] S61, the general information retrieval module receives the instruction from the global task management module "It is known that the biosynthetic pathway of L-lactic acid is: pyruvic acid + NADH + H + →L-lactic acid+NAD + , enzyme sequence, reaction conditions, substrate concentration, and detailed wet experiment operation steps are required to be planned", calling the large model interface reasoning;

[0106] S62, the general information retrieval module calls the search engine and the scientific research database to infer the wet experiment operation steps, including the reaction conditions, reagent volume and specific operations of each step. The global task management module is returned through the API. The global task management module determines that the general information retrieval module has completed the task, updates the task planning and execution description file, and schedules the wet experiment embodied execution module;

[0107] S7. Wet experiment program execution:

[0108] S71, the wet experiment embodied execution module receives the instruction from the global task management module: "The biosynthetic pathway of L-lactic acid is known to be: pyruvate + NADH + H + →L-lactic acid+NAD + , enzyme sequence, reaction conditions, substrate concentration and wet experiment operation steps, requiring the wet experiment to perform the above steps", calling the large model interface reasoning;

[0109] S72, wet experiment embodiment execution module generates control code:

[0110] S721, the module schedules the Opentrons-Flex liquid handling robot to configure the buffer solution corresponding to the preset pH, add the amount of pyruvate, NADH and LDH required by the step, and the temperature control module maintains a constant temperature at 25°C;

[0111] S722. After the reaction reaches the set time, the module controls the temperature of the reaction system to drop to 4°C to significantly slow down the reaction rate, controls the robotic arm to sample from the reaction system, and transmits the sample to the spectrometer to calculate the consumption rate of NADH by detecting the absorbance change of NADH at 340nm;

[0112] S73, the consumption rate of NADH is returned to the global task management module through the API, and the global task management module determines whether the wet experiment embodied execution module has completed the task based on the consumption rate: if the consumption rate of NADH is too low, the global task management module updates the task planning and execution description file (adjusts the wet experiment plan), such as increasing the injection rate of pyruvate or slightly increasing the temperature, and enters S71 to continue the cycle until the reaction rate meets the expectation, then it is determined that the wet experiment embodied execution module has completed the task, and the task planning and execution description file is updated;

[0113] S8. The user monitors the experimental progress and feedback information in real time through the front-end interface of the human-computer collaborative interaction module. The system provides the user with the current status of the experiment, including reaction conditions, NADH consumption rate, and experimental progress, through the WebSocket protocol. The user can enter new instructions or adjust experimental parameters based on the feedback.

[0114] The embodiment of the present invention provides a large model-driven embodied intelligent biosynthesis method, whose implementation principle and technical effects are the same as those of the aforementioned system embodiment. For the sake of brief description, for parts not mentioned in this embodiment, reference can be made to the corresponding contents in the aforementioned system embodiment, which will not be repeated here.

[0115] Further, refer to Figure 3 As shown, an embodiment of the present invention further provides an electronic device, which may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, and the processor executes the computer program to implement the operation of the above-mentioned system and the execution of the above-mentioned large model-driven embodied intelligent biosynthesis method.

[0116] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions of the electronic device and processes data by running or executing programs or modules stored in the memory 11, and calling data stored in the memory 11.

[0117] It should be understood by those skilled in the art that embodiments of the present invention may be provided as systems, methods, electronic devices or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0119] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0120] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A large model-driven embodied intelligent biosynthetic system, characterized in that: The system adopts a centralized multi-model intelligent agent architecture, which includes: human-computer collaborative interaction module, global task management module, reverse path planning module, protease modification module, metabolic pathway optimization module, general information retrieval module, wet experiment embodied execution module and code execution module; among them: The human-computer collaborative interaction module is used to receive the experimental requirements input by the user and transmit the experimental requirements to the global task management module; The global task management module is driven by a large model and is used to decompose experimental requirements into multiple tasks and schedule them to be executed by various functional modules; The reverse pathway planning module is used to infer the reaction precursors and search for substrate molecules based on the molecules to be synthesized input by the user, so as to generate the biosynthetic reaction pathway of the target product; The protease modification module selects the natural enzyme amino acid sequence based on the biosynthetic reaction pathway generated by the reverse pathway planning module, and generates a corresponding enzyme amino acid skeleton sequence modification scheme in combination with the synthesis target input by the user; The metabolic pathway optimization module optimizes the metabolic pathway in an in vitro or in vivo environment based on the generated biosynthetic reaction pathway and enzyme amino acid sequence; The general information retrieval module uses a large model-driven search engine to call external data sources to obtain biosynthesis knowledge and assist in generating wet experiment operation steps and plans; The wet experiment embodied execution module is used to convert the output wet experiment scheme into the control code of the embodied biosynthesis device, execute the wet experiment, and dynamically observe the experimental process and experimental results; The code execution module is used to call the code compiler interface to perform calculation tasks of simulating experimental conditions, analyzing experimental data, and optimizing experimental parameters; The human-computer collaborative interaction module, reverse path planning module, protease modification module, metabolic pathway optimization module, general information retrieval module, wet experiment embodied execution module and code execution module are all scheduled to start or end by the global task management module, and after the end, the running status and observation are fed back to the global task management module. The global task management module dynamically adjusts and optimizes the experimental plan according to the feedback information of each functional module, completes the overall planning and management of the task, and realizes the closed-loop optimization of dry and wet experiments of autonomous biosynthesis; The global task management module is developed based on the Python Flask or Django backend framework and integrates a large model API. It uses a combination of ReAct prompts and zero-sample thinking chain reasoning methods to dynamically call each functional module during the reasoning process. It decomposes the wet experiment demand instruction information sent by the human-computer collaborative interaction module and from the user into multiple executable subtasks. During the task execution process, it receives feedback information from other functional modules through the RESTAPI interface, and dynamically adjusts the task planning and decomposition plan accordingly. The planning and decomposition of tasks are managed through the task planning and execution description file in .json format, which records in detail the execution steps, dependencies, called tools and feedback mechanisms of the experiment. The decomposition of tasks and the calling of tools are scheduled through the task queue system, and the task status and feedback are managed through the message queue system. The wet experiment embodied execution module integrates the biosynthesis embodied experiment platform, the control document vector library and the large model API, wherein: The biosynthesis embodied experimental platform is used to separate the preparation of liquid systems from the analysis of liquid systems. The preparation of liquid systems uses an open source programmable liquid robot platform equipped with a pipette, a gripper, a heating and shaking module, a temperature control module, a magnetic control module, and a thermal cycle module; the analysis of liquid systems uses an ELISA reader, a spectrometer, and a chromatograph to monitor the experimental process in real time. The control document vector library stores the vector representation of the low-level and high-level control code development documents of each experimental device encoded by the multimodal large model; The large model combines input instructions with the retrieval-enhanced generation method to automatically generate control codes for the experimental platform and perform biological synthesis wet experiments; the synthesis and analysis results of the experiment are fed back to the global task management module through the REST API.

2. A large model-driven embodied intelligent biosynthesis system according to claim 1, characterized in that: The human-computer collaborative interaction module is also used to visualize the execution progress, task status and any changes of the solution according to the task planning and execution description files of the global task management module.

3. A large model-driven embodied intelligent biosynthesis system according to claim 1, characterized in that: The reverse pathway planning module uses a large model-driven enzymatic reaction database and a reverse pathway planning tool to infer and design the optimal metabolic pathway from the starting material to the target product.

4. A large model-driven embodied intelligent biosynthesis system according to claim 1, characterized in that: The protease modification module screens and optimizes key enzymes in biosynthesis based on protein language models, structure prediction tools and enzyme kinetics simulation tools.

5. A large model-driven embodied intelligent biosynthesis system according to claim 1, characterized in that: The metabolic pathway optimization module optimizes and transforms the metabolic network of the chassis in an in vivo environment in combination with the genome-scale metabolic network according to the inherent metabolic environment of the microbial chassis.

6. A large model-driven embodied intelligent biosynthesis system according to claim 1, characterized in that: The general information retrieval module integrates a large model API, an external search engine, and an internal vector database to retrieve experimental data and literature that are highly relevant to the current biosynthesis task based on semantic similarity.

7. A large model-driven embodied intelligent biosynthesis system according to claim 1, characterized in that: The code execution module receives task codes from other functional modules through the REST API interface. The task codes are passed into the Docker container for execution. After the execution is completed, the results are fed back to the global task management module through the REST API.

8. A large model-driven embodied intelligent biosynthesis method, characterized in that: Using the large model-driven embodied intelligent biosynthesis system as described in any one of claims 1 to 7 to achieve autonomous biosynthesis, the method comprising: Receive the experimental requirements input by the user through the human-computer collaborative interaction module, and transmit the experimental requirements to the global task management module; Use the global task management module to decompose the experimental requirements into multiple tasks, dispatch them to each functional module for execution, and dynamically adjust and optimize the experimental plan based on the feedback information of each functional module to complete the overall planning and management of the task; The reverse pathway planning module is used to infer the reaction precursors and search for substrate molecules based on the molecules to be synthesized input by the user, thus generating the biosynthetic reaction pathway of the target product; The protease modification module selects the natural enzyme amino acid sequence based on the biosynthetic reaction pathway and generates the corresponding enzyme amino acid skeleton sequence modification scheme in combination with the synthesis target input by the user; The metabolic pathway optimization module optimizes metabolic pathways in vitro or in vivo based on the generated biosynthetic reaction pathways and enzyme amino acid sequences; Through the general information retrieval module, a large model-driven search engine is used to call external data sources to obtain biosynthetic knowledge and assist in generating wet experiment operation steps and plans; The wet experiment embodied execution module is used to convert the output wet experiment plan into the control code of the embodied biosynthesis equipment, execute the wet experiment, and dynamically observe the experimental process and experimental results; The code execution module is used to call the code compiler interface to perform computational tasks such as simulating experimental conditions, analyzing experimental data, and optimizing experimental parameters.

Images