Artificial intelligence chemical inverse synthesis route planning system

Abstract

The invention discloses an artificial intelligence chemical inverse synthesis route planning system, and belongs to the field of intelligent planning systems, the method comprises the following steps: a user interface, a Monte Carlo tree search and route search module, a reaction condition prediction and route optimization module, and an output and feedback module; wherein the user interface is used for providing a customized interface for a user, and the user performs molecule input based on the customized interface to determine a target molecule; the Monte Carlo tree search and path search module is used for constructing an inverse synthesis path of a target molecule based on a Monte Carlo tree search algorithm; the reaction condition prediction and path optimization module is used for predicting the reaction condition of each candidate reaction and optimizing the inverse synthesis path of the generated target molecule to obtain an optimal inverse synthesis path; and the output and feedback module is used for displaying the optimal inverse synthesis path to a user and receiving feedback of the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent planning systems, and in particular relates to an artificial intelligence chemical retrosynthesis route planning system. Background Art

[0002] With the continuous development of chemical synthesis, retrosynthetic pathway planning plays a vital role in drug development, materials science and organic chemistry research. Traditional retrosynthetic pathway planning relies on the expertise and experience of chemists to design synthetic routes through manual analysis and literature retrieval. This method is time-consuming and inefficient, especially when faced with complex molecules or novel compounds, where chemists' knowledge and experience may be limited, making it difficult to quickly find the optimal synthetic pathway.

[0003] In recent years, with the rapid development of artificial intelligence technology, especially the application of machine learning and deep learning technology in the field of chemistry, the method of retrosynthetic path planning has been revolutionized. AI technology can automatically predict the retrosynthetic path and reaction conditions of chemical reactions by learning a large amount of chemical reaction data, greatly improving the efficiency and accuracy of synthetic path design. Among them, the Monte Carlo tree search algorithm (MCTS) combined with the method of deep neural network has shown great potential in retrosynthetic path planning. However, the existing technology still has some limitations, such as insufficient generalization ability for specific reaction types, and problems such as inaccurate predicted reaction conditions. Therefore, the present invention provides an artificial intelligence chemical retrosynthetic route planning system. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an artificial intelligence chemical retrosynthesis route planning system to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above objectives, the present invention provides an artificial intelligence chemical retrosynthesis route planning system, comprising:

[0006] User interface, Monte Carlo tree search and path search module, reaction condition prediction and path optimization module, output and feedback module;

[0007] Wherein, the user interface is used to provide a customized interface for the user, and the user inputs molecules based on the customized interface to determine the target molecule;

[0008] The Monte Carlo tree search and path search module is used to construct a retrosynthetic path of the target molecule based on the Monte Carlo tree search algorithm;

[0009] The reaction condition prediction and pathway optimization module is used to predict the reaction conditions of each candidate reaction and optimize the retrosynthetic pathway of the generated target molecule to obtain the best retrosynthetic pathway;

[0010] The output and feedback module is used to display the optimal retrosynthetic pathway to the user and receive feedback from the user.

[0011] Optionally, the user interface includes: a graphical user interface, a molecular analysis unit and a multi-language switching unit;

[0012] Wherein, the graphical user interface is used to provide an analysis input interface for the user;

[0013] The molecular analysis unit is used to analyze the molecules uploaded by the user to generate internal representations of the molecules;

[0014] The multi-language switching unit is used for the user to switch the language mode.

[0015] Optionally, the Monte Carlo tree search and path search module includes: a Monte Carlo tree construction unit, a function selection unit and an expansion unit;

[0016] The Monte Carlo tree construction unit is used to construct a Monte Carlo tree with the target molecule as the root node, and assign an initial score and state to each node;

[0017] The function selection unit is used to evaluate and select the most potential leaf node based on the UCB algorithm;

[0018] The expansion unit is used to expand the selected leaf node with the greatest potential.

[0019] Optionally, the reaction condition prediction and path optimization module includes a reaction condition prediction unit, a path evaluation unit and a path optimization unit;

[0020] The reaction condition prediction unit is used to predict the optimal conditions of the chemical reaction based on the expanded leaf node, and combine the predicted optimal reaction conditions with the reaction template to generate a retrosynthetic pathway with specific reaction conditions;

[0021] The path evaluation unit is used to evaluate the expanded leaf node and update the evaluation result to the score of the corresponding node in the Monte Carlo tree;

[0022] The path optimization unit is used to repeatedly perform function selection, leaf node expansion and path evaluation until an optimal path score is obtained.

[0023] Optionally, the output and feedback module includes a result output unit, a user evaluation unit and a system self-learning unit;

[0024] Wherein, the result output unit is used to output the optimal retrosynthetic pathway and present the result to the user in a graphical manner;

[0025] The user evaluation unit is used to evaluate the output result and give evaluation opinions;

[0026] The system self-learning unit is used for the system to perform self-learning and optimization based on the user's evaluation opinions.

[0027] Optionally, the artificial intelligence chemical retrosynthesis route planning system further includes: a system integration and expansion module, a hardware optimization and performance monitoring module;

[0028] Wherein, the system integration and expansion module is used to remotely communicate with other chemical information management systems based on a network interface;

[0029] The hardware optimization and performance monitoring module is used to track and record user operations and monitor operating performance.

[0030] The present invention provides a computer terminal device, comprising:

[0031] one or more processors;

[0032] A memory, coupled to the processor, for storing one or more programs;

[0033] When the one or more programs are executed by the one or more processors, the one or more processors implement an artificial intelligence chemical retrosynthesis route planning system.

[0034] The present invention provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, an artificial intelligence chemical retrosynthesis route planning system is implemented.

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] The present invention provides a comprehensive retrosynthetic pathway planning solution, which utilizes a Monte Carlo tree search algorithm and deep learning technology to achieve rapid and accurate prediction of the retrosynthetic pathway of chemical molecules.

[0037] The present invention not only improves the efficiency and accuracy of retrosynthetic pathway planning, but also provides users with a comprehensive, flexible, and user-friendly tool to support synthetic pathway design in chemical research and industrial applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0039] Figure 1 A step diagram for developing a chemical retrosynthesis system according to an embodiment of the present invention;

[0040] Figure 2The reverse synthetic route of branebrutinib (BMS-986195) of the present invention;

[0041] Figure 3 Different representations of simple esterification reactions according to embodiments of the present invention;

[0042] Figure 4 Predicting reactants based on templates, sequences, and graphs according to an embodiment of the present invention;

[0043] Figure 5 It is a diagram of a synthesis plan based on Monte Carlo tree search according to an embodiment of the present invention;

[0044] Figure 6 A schematic diagram of the entire process of developing a chemical retrosynthesis system based on an experience-guided network and Monte Carlo tree search according to an embodiment of the present invention;

[0045] Figure 7 This is the prediction result of the retrosynthesis of the molecule according to the embodiment of the present invention. DETAILED DESCRIPTION

[0046] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0048] Embodiment 1

[0049] The present invention aims to provide an improved retrosynthetic pathway planning method, which can more accurately predict the retrosynthetic pathway and reaction conditions of chemical reactions, and has the characteristics of high efficiency, flexibility and user-friendliness. By combining the Monte Carlo tree search algorithm and deep learning technology, the present invention can not only handle complex chemical synthesis problems, but also provide practical reaction condition predictions, thereby providing chemists and researchers with a powerful synthetic pathway design tool.

[0050] like Figure 1 As shown, this embodiment provides an artificial intelligence chemical retrosynthesis route planning system, which uses a Monte Carlo tree search algorithm to simulate the retrosynthesis path of a generated molecule and combines a deep neural network to predict reaction conditions, including the following steps:

[0051] S1. Input the chemical structure information of the target molecule, including the SMILES representation, molecular graph or other chemical structure representation of the molecule.

[0052] S2. Initialize the Monte Carlo tree, set the target molecule as the root node of the tree, and define the node structure in the tree, including molecular representation, reaction conditions, path scores, etc.

[0053] S3. Use the upper confidence bound statistic (UCB) as the selection function to select the most potential nodes from the leaf nodes of the Monte Carlo tree for expansion.

[0054] S4. The selected leaf nodes are expanded by applying a trained neural network to predict possible reaction centers and breakable bonds, generating a set of candidate precursor molecules and corresponding reaction conditions.

[0055] S5. Evaluate each candidate precursor molecule to determine whether it is a purchasable precursor or a synthesizable intermediate, and update the node score based on the evaluation result.

[0056] S6. Repeat steps S3 to S5 until a predetermined number of iterations or time limit is met, or a satisfying retrosynthetic pathway is found.

[0057] S7. Starting from the root node, a path is selected according to the node score to construct the retrosynthetic path of the target molecule, and the path details are output, including the synthesis steps, reaction conditions, required raw materials, and expected products.

[0058] This embodiment also includes a computer system for retrosynthesis path planning, which includes at least one processor; a memory for storing input and output devices of the chemical synthesis reaction condition prediction device and corresponding software programs; an input and output device for interacting with a user and providing structural information of the target molecule; and a processor for executing the chemical synthesis reaction condition prediction method.

[0059] This embodiment also provides a chemical synthesis reaction condition prediction device, which can predict the optimal conditions of a chemical reaction according to the structural characteristics of a target molecule, including catalyst, temperature and pressure. The device includes: an input module for receiving the structural information of the target molecule; a prediction module for predicting the optimal conditions of a chemical reaction; a storage device for storing a large amount of chemical reaction data required for training a neural network model; an output module for outputting the predicted reaction conditions and reverse synthesis path and providing a user interaction function.

[0060] The chemical synthesis reaction condition prediction method applied to the chemical synthesis reaction condition prediction device includes: using a neural network model to analyze the structural characteristics of the target molecule; predicting the optimal conditions for the chemical reaction, including catalyst, temperature and pressure, based on the analysis results; applying the predicted reaction conditions to a chemical reaction template to generate a retrosynthetic path with specific reaction conditions.

[0061] This embodiment also includes a chemical synthesis database for storing and retrieving the retrosynthetic pathways and corresponding reaction conditions generated by the method of the present invention, so that users can quickly find and use existing synthesis strategies.

[0062] An artificial intelligence chemical retrosynthesis route planning system includes: a user interface, a Monte Carlo tree search and path search module, a reaction condition prediction and path optimization module, and an output and feedback module; wherein the user interface is used to provide a customized interface for users, and the user determines the target molecule by molecular input based on the customized interface; the Monte Carlo tree search and path search module is used to construct the retrosynthesis path of the target molecule based on the Monte Carlo tree search algorithm; the reaction condition prediction and path optimization module is used to predict the reaction conditions of each candidate reaction and optimize the retrosynthesis path of the generated target molecule to obtain the best retrosynthesis path; the output and feedback module is used to display the best retrosynthesis path to the user and receive feedback from the user.

[0063] User interface operation and molecule input

[0064] The user interface includes: a graphical user interface, a molecular analysis unit and a multi-language switching unit; wherein the graphical user interface is used to provide the user with an analysis input interface; the molecular analysis unit is used to generate an internal representation of the molecule from the molecular analysis uploaded by the user; and the multi-language switching unit is used for the user to switch the language mode. The specific implementation of this embodiment includes:

[0065] (1) The user opens a chemical synthesis software application and enters the retrosynthetic pathway planning module through a graphical user interface (GUI).

[0066] (2) The user uploads the target molecule file (such as Mol file) through the GUI or directly draws the structure of the target molecule on the interface.

[0067] (3) The system parses the uploaded file or drawn structure and generates an internal representation of the molecule, such as a molecular map or SMILES string.

[0068] (4) After the user confirms that the structure of the target molecule is correct, click the "Start Planning" button to start the retrosynthetic pathway planning process.

[0069] (5) The system can provide a multilingual user interface to support users from different countries and regions.

[0070] (6) The system can provide customized interfaces and functions to meet the special needs of specific user groups or specific chemical research fields.

[0071] Monte Carlo Tree Search and Path Prediction

[0072] The Monte Carlo tree search and path search module includes: a Monte Carlo tree construction unit, a function selection unit and an expansion unit; wherein the Monte Carlo tree construction unit is used to construct a Monte Carlo tree with the target molecule as the root node and assign an initial score and state to each node; the function selection unit is used to evaluate and select the most potential leaf node based on the UCB algorithm; the expansion unit is used to expand the selected most potential leaf node. The specific implementation of this embodiment includes:

[0073] (1) The system initializes the Monte Carlo tree and sets the parsed target molecule as the root node of the tree.

[0074] (2) The system executes the selection function and selects the leaf node with the most potential for expansion based on the UCB algorithm.

[0075] (3) For the selected leaf nodes, the system applies a well-trained deep neural network to predict possible reaction centers and breakable bonds, generating a set of candidate precursor molecules and reaction conditions.

[0076] (4) The system evaluates each candidate precursor molecule to determine whether it is a purchasable precursor or a synthesizable intermediate and updates the score of the corresponding node.

[0077] (5) The system repeats the selection, expansion, and evaluation steps until a predetermined number of iterations or a time limit is reached.

[0078] (6) The system starts from the root node, selects the optimal path based on the node score, and constructs the retrosynthetic pathway of the target molecule.

[0079] (7) The system includes an algorithm library containing a variety of prediction algorithms and models for selecting the most appropriate prediction tool based on different chemical synthesis requirements.

[0080] (8) The system includes a reaction database for storing and retrieving various chemical reaction templates and known synthetic pathways so that users can select and modify existing synthetic strategies.

[0081] (9) The system includes a compound library for storing and retrieving structural information and property data of various compounds so that users can quickly find and use common chemical reagents and intermediates.

[0082] (10) The system includes a simulation environment for simulating chemical reaction processes and visualizing retrosynthetic pathways so that users can intuitively understand the synthesis process and predict the results.

[0083] (11) The system includes an optimization engine that automatically adjusts and optimizes each step in the retrosynthetic pathway to improve synthesis efficiency and product quality.

[0084] Reaction condition prediction and path optimization

[0085] The reaction condition prediction and path optimization module includes a reaction condition prediction unit, a path evaluation unit and a path optimization unit; wherein the reaction condition prediction unit is used to predict the optimal conditions of the chemical reaction based on the extended leaf node, and combines the predicted optimal reaction conditions with the reaction template to generate a retrosynthetic path with specific reaction conditions; the path evaluation unit is used to evaluate the extended leaf node and update the evaluation result to the score of the corresponding node in the Monte Carlo tree; the path optimization unit is used to repeatedly perform function selection, leaf node expansion and path evaluation until the optimal path score is obtained. The specific implementation of this embodiment includes:

[0086] (1) In the expansion step, the system uses a neural network to predict the reaction conditions for each candidate reaction, including catalyst, temperature, pressure, and solvent.

[0087] (2) The system combines the predicted reaction conditions with the reaction template to form a complete chemical reaction scheme and evaluates the feasibility of the scheme.

[0088] (3) The system optimizes the generated retrosynthetic pathways and selects the best pathway based on the feasibility of the reaction conditions and the availability of the precursor molecules.

[0089] Result output and user feedback

[0090] The output and feedback module includes a result output unit, a user evaluation unit and a system self-learning unit; wherein the result output unit is used to output the optimal retrosynthetic path and display the result to the user in a graphical manner; the user evaluation unit is used to evaluate the output result and give evaluation opinions; the system self-learning unit is used for the system to self-learn and optimize based on the user's evaluation opinions. The specific implementation methods of this embodiment include:

[0091] (1) After the system completes the retrosynthetic pathway planning, the results are displayed to the user in a graphical manner, including the synthesis steps, reaction conditions, and required raw materials.

[0092] (2) Users can view detailed information for each step, including reaction equation, raw material sources, and estimated costs.

[0093] (3) The user evaluates the prediction results and provides comments and suggestions through the user feedback module. The system includes an environmental impact assessment module for evaluating the potential impact of the predicted retrosynthetic pathway on the environment and providing environmentally friendly synthesis suggestions. The system includes a safety assessment module for evaluating the potential safety risks in the predicted retrosynthetic pathway and providing corresponding safety suggestions and preventive measures. The system includes a cost analysis module for calculating the expected cost of the predicted retrosynthetic pathway and providing a cost-benefit analysis report.

[0094] (4) The system conducts self-learning and optimization based on user feedback, adjusts algorithm parameters, and improves prediction accuracy. The system includes a user feedback module to collect users' evaluation and suggestions on prediction results in order to continuously improve and upgrade system performance.

[0095] System integration and expansion

[0096] (1) Users hope to integrate the system of the present invention with the existing chemical information management system (CIMS) to achieve data sharing and process automation.

[0097] (2) The system provides API interfaces and integration guides to help users integrate the retrosynthesis pathway planning module with the CIMS system.

[0098] (3) Users call the retrosynthesis pathway planning service through the API and feed the results back to the CIMS to achieve seamless data connection. The system can communicate with remote users or other systems through a network interface to provide retrosynthesis pathway planning and reaction condition prediction services.

[0099] (4) The system includes a patent search module for retrieving relevant chemical synthesis patent information to avoid infringement of existing intellectual property rights.

[0100] Hardware optimization and performance monitoring

[0101] (1) To ensure system performance under high load, system administrators monitor server performance, including CPU usage, memory usage, and response time.

[0102] (2) The system provides performance monitoring tools to track the efficiency and resource consumption of algorithm execution in real time. The system includes logging capabilities to track and record user operations and system events to facilitate problem diagnosis and performance monitoring.

[0103] (3) Based on the monitoring results, the administrator tunes the system, such as increasing computing resources, optimizing algorithm implementation, or adjusting concurrency strategies, to improve system performance.

[0104] (4) The system includes data backup and recovery functions to protect the stored retrosynthetic pathway and reaction condition data from accidental loss or damage.

[0105] Embodiment 2

[0106] The process of retrosynthetic design: (A) Process the reactions in the dataset by adding atomic mappings of reactants and products; (B) Extract transformation rules from reactions with atomic mappings to guide the computer in retrosynthesis; (C) Iteratively predict and rank the precursors according to the indicators of the evaluation route until the end condition is reached; (D) There are two situations for the route generated by the model: success and failure.

[0107] A retrosynthetic path planning method comprises the following steps:

[0108] Target molecule input and parsing: Users input the chemical structure information of the target molecule through a graphical user interface (GUI) or a programming interface (API). The system supports multiple molecular representation formats such as SMILES strings, InChI keys, Mol files, and molecular maps. The system contains a parsing engine that converts the input chemical structure into an internal representation for further processing.

[0109] Monte Carlo tree construction and initialization: The system initializes the Monte Carlo tree, takes the target molecule as the root node, and assigns an initial score and state to each node. The initial score of the node is based on the synthetic and commercial availability of the molecule.

[0110] Selection function: Implement a selection function that uses the UCB algorithm to evaluate and select the most promising leaf nodes for expansion. The UCB algorithm considers the node's visit frequency, score, and exploration value to balance exploration and exploitation.

[0111] Extension strategy: For the selected leaf nodes, a deep neural network is applied to predict possible reaction centers, broken bonds, and reaction conditions. The neural network is trained based on a large amount of known chemical reaction data and can identify reaction templates and reaction mechanisms.

[0112] Reaction Condition Prediction: Neural networks predict conditions related to chemical reactions, such as catalysts, temperature, pressure, and solvents. Reaction condition predictions are combined with reaction templates to form complete chemical reaction schemes.

[0113] Path evaluation and feedback: Evaluate the generated precursor molecules to determine whether they are purchasable precursors or synthesizable intermediates. The evaluation results are used to update the scores of the corresponding nodes in the Monte Carlo tree, providing a basis for subsequent path selection.

[0114] Iterative optimization and pathway selection: The system repeatedly performs the selection, expansion, and evaluation steps until a predetermined number of iterations or time limit is met. Each iteration aims to optimize the pathway score and improve the feasibility and efficiency of the retrosynthetic pathway.

[0115] Pathway construction and output: Starting from the root node, select the optimal path according to the node score to construct the retrosynthetic path of the target molecule. Output a detailed retrosynthetic path report, including synthesis steps, reaction conditions, required raw materials and expected products.

[0116] Computer system implementation: Develop a computer system, including processors, memory, input and output devices, and software programs, to implement the above solutions. The system can run on multiple operating systems, such as Windows, Linux, and macOS.

[0117] Database integration: Build a chemical synthesis database to store and retrieve retrosynthetic pathways and reaction conditions. The database supports fast retrieval to help users find similar synthesis cases and reaction templates.

[0118] User interface design: Design an intuitive user interface that allows users to easily input molecular structures, select reaction conditions, and view predicted results. Provide interactive charts and visualization tools to help users understand complex chemical pathways.

[0119] Algorithm library and model management: Integrate multiple prediction algorithms and models to provide users with the most appropriate prediction tools. The algorithm library is regularly updated to include the latest chemical research results and prediction technologies.

[0120] Safety and Compliance Assessment: Includes a safety assessment module to evaluate potential safety risks in predicted pathways. Compliance checks are provided to ensure that synthetic pathways meet current chemical safety standards and regulatory requirements.

[0121] Cost-benefit analysis: Implement a cost analysis module to estimate the raw material cost, production cost and expected benefits of the predicted pathway, and provide users with cost-effective synthetic pathway recommendations.

[0122] Environmental Impact Assessment: Integrated environmental impact assessment tools to analyze the potential impact of synthetic pathways on the environment. Provide environmentally friendly synthesis suggestions to reduce the burden of chemical synthesis on the environment.

[0123] Patent search and avoidance: Provide patent search functions to help users identify and avoid potential intellectual property conflicts. Provide users with patent avoidance strategies to ensure the innovation and legality of the synthesis path.

[0124] User feedback and system optimization: including user feedback module, collecting users’ comments and suggestions on prediction results. Based on user feedback and system performance data, continuously optimize the algorithm and user interface.

[0125] Multi-language and international support: Provide multi-language user interface to support users from different countries and regions. Consider cultural differences and region-specific chemical naming conventions to provide a localized user experience.

[0126] Customized services and API: Provide customized services to meet the special needs of specific user groups or research fields. Open API allows third-party developers to integrate and expand system functions.

[0127] Advanced Analysis Tools: Includes an advanced analysis toolset for in-depth analysis of predicted retrosynthetic pathways, such as pathway length, synthetic complexity, raw material availability, etc. A customizable report generator is provided, allowing users to generate detailed synthesis reports as needed.

[0128] Education and Training Resources: Provides online education and training resources to help users understand the basics and best practices of retrosynthetic pathway planning. Includes tutorials, case studies, FAQs, and webinars.

[0129] Technical support and maintenance: Provide professional technical support and maintenance services to ensure the system runs stably and meets user needs. Regularly update the system, fix known problems and introduce new features.

[0130] Data backup and recovery: Implement data backup and recovery functions to ensure the security and reliability of user data. Provide data backup strategies and recovery guides to help users quickly recover data when it is lost or damaged.

[0131] System integration and expansion: Provide system integration tools and guidelines to facilitate users to integrate the system of the present invention with other chemical information management systems (CIMS) or enterprise resource planning (ERP) systems. Support system expansion, allowing users to add new functional modules or algorithms as needed.

[0132] Hardware optimization and compatibility: Optimize system support for different hardware platforms, including personal computers, servers, and cloud computing environments. Ensure that the system can maintain high performance and stability under a variety of hardware configurations.

[0133] Performance monitoring and tuning: Implement performance monitoring tools to track system operation status and resource usage in real time. Provide performance tuning guides to help users optimize based on system load and performance requirements.

[0134] User access control and permission management: Implement user access control and permission management functions to ensure the security of data and functions. Provide flexible permission setting options, allowing administrators to assign access rights based on user roles and needs.

[0135] Multi-factor authentication and data encryption: Support multi-factor authentication (MFA) to enhance system security. Implement data transmission and storage encryption to protect the privacy and integrity of user data.

[0136] like Figure 2Figure 1 shows the retrosynthetic route of branebrutinib (BMS-986195). It starts with the target molecule (green box) and ends with commercially available reactants (yellow box). The yellow and gray circles highlight the reaction sites of the current and previous steps, respectively. The "$" indicates the market price of the molecule in grams.

[0137] like Figure 3 Figure 1 shows different representations of a simple esterification reaction. In the graphical reaction representation, the red, black, and gray balls represent oxygen, carbon, and hydrogen atoms, respectively. The yellow numbers on the balls indicate the number of atoms mapped.

[0138] Each chemical reaction is typically represented by a set of atom maps, where each atom map is a graph with a set of discontinuous edges that constructs a one-to-one correspondence between all atoms in reactants and products. In addition, atom maps are often used to separate reactants and reagents in a reaction. The only reaction representation that is independent of atom maps is one that does not distinguish between reactants and reagents. Accurate atom maps can facilitate downstream tasks, such as determining the efficiency of a transformation pathway by counting the number of conserved carbon atoms in a reaction, and can also be used to track atoms to understand reaction mechanisms.

[0139] like Figure 4 Figure 1 shows the basic workflow of template-based and template-independent retrosynthetic methods. (A) Template-based method - products are converted to reactants based on templates in the template library. (B) Sequence-based method - the model translates the SMILES string of the product molecule into the SMILES string of the reactant. (C) Graph-based method - the bonds of the product are separated into synthons, and the synthons are completed to form reactants.

[0140] like Figure 5-Figure 6 As shown, a Monte Carlo search tree (MCTS) is used to search for the retrosynthetic path, and the expansion of the search tree is carried out through four iterative steps, including selection, expansion, simulation and backpropagation.

[0141] Selection: Traverse the search tree from the root node to the leaf node by selecting the child node with the largest upper confidence bound (UCB) score.

[0142] Extension: Create child nodes by sampling from chemical retrosynthetic models.

[0143] Simulation: Evaluate each search position in state space, find the best position, and then search from this position until the goal is found to create a path to the terminal node.

[0144] Back propagation: Calculate the rewards of the terminal nodes and update the UCB scores of the upper nodes.

[0145] like Figure 7As shown in FIG, a neural network-based method is taken as an example: the EG-MCTS method consists of two stages, namely (I) learning an experience-guided network (EGN) for guiding the search, and (II) using the learned EGN to generate a synthetic route for the molecule.

[0146] Through the above solutions, the present invention not only improves the efficiency and accuracy of retrosynthetic pathway planning, but also provides users with a comprehensive, flexible, and user-friendly tool to support synthetic pathway design in chemical research and industrial applications.

[0147] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An artificial intelligence chemical retrosynthesis route planning system, characterized in that: include: User interface, Monte Carlo tree search and path search module, reaction condition prediction and path optimization module, output and feedback module; Wherein, the user interface is used to provide a customized interface for the user, and the user inputs molecules based on the customized interface to determine the target molecule; The Monte Carlo tree search and path search module is used to construct a retrosynthetic path of the target molecule based on the Monte Carlo tree search algorithm; The reaction condition prediction and pathway optimization module is used to predict the reaction conditions of each candidate reaction and optimize the retrosynthetic pathway of the generated target molecule to obtain the best retrosynthetic pathway; The output and feedback module is used to display the optimal retrosynthetic pathway to the user and receive feedback from the user.

2. The artificial intelligence chemical retrosynthesis route planning system according to claim 1, characterized in that: The user interface includes: a graphical user interface, a molecular analysis unit and a multi-language switching unit; Wherein, the graphical user interface is used to provide an analysis input interface for the user; The molecular analysis unit is used to analyze the molecules uploaded by the user to generate internal representations of the molecules; The multi-language switching unit is used for the user to switch the language mode.

3. The artificial intelligence chemical retrosynthesis route planning system according to claim 1, characterized in that: The Monte Carlo tree search and path search module includes: a Monte Carlo tree construction unit, a function selection unit and an expansion unit; The Monte Carlo tree construction unit is used to construct a Monte Carlo tree with the target molecule as the root node, and assign an initial score and state to each node; The function selection unit is used to evaluate and select the most potential leaf node based on the UCB algorithm; The expansion unit is used to expand the selected leaf node with the greatest potential.

4. The artificial intelligence chemical retrosynthesis route planning system according to claim 3, characterized in that: The reaction condition prediction and path optimization module includes a reaction condition prediction unit, a path evaluation unit and a path optimization unit; The reaction condition prediction unit is used to predict the optimal conditions of the chemical reaction based on the expanded leaf node, and combine the predicted optimal reaction conditions with the reaction template to generate a retrosynthetic pathway with specific reaction conditions; The path evaluation unit is used to evaluate the expanded leaf node and update the evaluation result to the score of the corresponding node in the Monte Carlo tree; The path optimization unit is used to repeatedly perform function selection, leaf node expansion and path evaluation until an optimal path score is obtained.

5. The artificial intelligence chemical retrosynthesis route planning system according to claim 1, characterized in that: The output and feedback module includes a result output unit, a user evaluation unit and a system self-learning unit; Wherein, the result output unit is used to output the optimal retrosynthetic pathway and present the result to the user in a graphical manner; The user evaluation unit is used to evaluate the output result and give evaluation opinions; The system self-learning unit is used for the system to perform self-learning and optimization based on the user's evaluation opinions.

6. The artificial intelligence chemical retrosynthesis route planning system according to claim 1, characterized in that: The artificial intelligence chemical retrosynthesis route planning system also includes: a system integration and expansion module, a hardware optimization and performance monitoring module; Wherein, the system integration and expansion module is used to remotely communicate with other chemical information management systems based on a network interface; The hardware optimization and performance monitoring module is used to track and record user operations and monitor operating performance.

7. A computer terminal device, characterized in that: include: one or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the artificial intelligence chemical retrosynthesis route planning system as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the artificial intelligence chemical retrosynthesis route planning system as described in any one of claims 1 to 6 is implemented.

Images