A retrosynthetic analysis method and device based on search experience

Abstract

A retrosynthesis analysis method and device based on search experience, the method first constructs a reverse reaction template library, then constructs a SoftMax neural network model, trains the SoftMax neural network model through a cross entropy loss function and the reverse reaction template library; jointly generates a route search algorithm with the trained SoftMax neural network model and a tree search algorithm and performs an iterative route search, sets labels for all reverse reaction templates along the reaction route, and obtains an empirical data set; constructs an empirical model, and trains the empirical model through an empirical data set and a binary cross entropy loss function; adjusts the output probability of the empirical model by a probability distribution based on the Bayesian theorem, and obtains a posterior probability distribution of the reverse reaction template; generates a pre-reactant for the product through a route search algorithm according to the posterior probability distribution of the reverse reaction template. The present invention has multi-step synthesizable information, which can greatly improve the algorithm's solving ability.

Description

Technical Field

[0001] The invention belongs to the technical field of compound synthesis analysis, and in particular relates to a retrosynthetic analysis method and device based on search experience. Background Art

[0002] Retrosynthetic analysis is a method used by chemists or computers to synthesize a given compound by breaking it down step by step into simpler reactants until a purchasable building block is found.

[0003] At present, the industry usually relies too much on the use of "static models" in retrosynthetic route planning, that is, using a single-step model trained on a specified data set combined with a tree search algorithm to plan the route. Since the data set often only contains single-step reactions and lacks multi-step synthesizable information, the algorithm's solving ability (that is, the decomposed building blocks can all be purchased) is insufficient.

[0004] Therefore, inventing a method with multi-step synthesizable information to improve the algorithm's solving ability has become an urgent problem to be solved. Summary of the invention

[0005] To this end, the present invention provides a retrosynthetic analysis method and device based on search experience, which makes up for this defect by using an empirical model trained by empirical data collected during route planning. The empirical model learns multi-step synthesizable information, which complements the single-step reaction information learned by the single-step model. The combination of the two can greatly improve the algorithm's solving ability.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a retrosynthetic analysis method based on search experience, comprising:

[0007] Constructing a reverse reaction template library, wherein the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field;

[0008] Constructing a SoftMax neural network model, and training the SoftMax neural network model through a cross entropy loss function and the inverse reaction template library;

[0009] The trained SoftMax neural network model is combined with a tree search algorithm to generate a route search algorithm; an iterative route search is performed by the route search algorithm, and labels are set for all reverse reaction templates along the reaction route to obtain an empirical data set; the empirical data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field;

[0010] Constructing an empirical model, and training the empirical model using the empirical data set and a binary cross entropy loss function;

[0011] The output probability of the empirical model is adjusted for probability distribution based on Bayes' theorem to obtain the posterior probability distribution of the reverse reaction template; and the product is generated as a pre-reactant through the route search algorithm according to the posterior probability distribution of the reverse reaction template.

[0012] As a preferred scheme of the retrosynthetic analysis method based on search experience, the SoftMax neural network model is obtained by constructing a fully connected multi-classification neural network model, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the vectorized representation data of the product is input into the SoftMax neural network model, processed by the SoftMax neural network model, and the probability distribution of each reverse reaction template in the reverse reaction template library is output.

[0013] 6. As a preferred embodiment of the retrosynthetic analysis method based on search experience, during the generation of the empirical data set, a target compound library is constructed, and the target compound library includes a compound SMILES field; after the iterative route search is completed by the route search algorithm, the label of all the reverse reaction templates along the solved reaction route is set to 1, and the label of all the reverse reaction templates along the unsolved reaction route is set to 0;

[0014] The empirical data set is obtained by searching the target compound library for empirical data collection.

[0015] As a preferred scheme of a retrosynthetic analysis method based on search experience, the empirical model is obtained by constructing a fully connected binary classification neural network model, and the empirical model is trained by the empirical data set and the binary cross entropy loss function; the vectorized representation data of the product and the vectorized representation data of the reverse reaction template are respectively input into the empirical model, processed by the empirical model, and the probability that the product can be solved by the reaction route after being processed by the reverse reaction template is output.

[0016] As a preferred scheme of the retrosynthetic analysis method based on search experience, in the expansion stage of the route search algorithm, the probability distribution of each reverse reaction template in the reverse reaction template library output by the SoftMax neural network model is used as the prior probability distribution of the reverse reaction template.

[0017] The present invention also provides a retrosynthetic analysis processing device based on search experience, which adopts the above-mentioned retrosynthetic analysis method based on search experience, comprising:

[0018] A reverse reaction template library construction module, used to construct a reverse reaction template library, wherein the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field;

[0019] A SoftMax neural network model construction and training module is used to construct a SoftMax neural network model and train the SoftMax neural network model through a cross entropy loss function and the inverse reaction template library;

[0020] An experience data set acquisition module is used to combine the trained SoftMax neural network model with the tree search algorithm to generate a route search algorithm; perform iterative route search through the route search algorithm, set labels for all reverse reaction templates along the reaction route, and obtain an experience data set; the experience data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field;

[0021] An empirical model construction and training module, used to construct an empirical model and train the empirical model using the empirical data set and a binary cross entropy loss function;

[0022] The pre-reactant generation module is used to adjust the output probability of the empirical model based on the probability distribution of Bayes' theorem to obtain the posterior probability distribution of the reverse reaction template; and generate the pre-reactant for the product through the route search algorithm according to the posterior probability distribution of the reverse reaction template.

[0023] As a preferred solution of a retrosynthetic analysis and processing device based on search experience, in the SoftMax neural network model construction and training module, the SoftMax neural network model is obtained by constructing a fully connected multi-classification neural network model, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the vectorized representation data of the product is input into the SoftMax neural network model, processed by the SoftMax neural network model, and the probability distribution of each reverse reaction template in the reverse reaction template library is output.

[0024] As a preferred solution of a retrosynthetic analysis processing device based on search experience, in the experience data set acquisition module, during the generation of the experience data set, a target compound library is constructed, and the target compound library includes a compound SMILES field;

[0025] After the iterative route search is completed by the route search algorithm, the labels of all the reverse reaction templates along the solved reaction route are set to 1, and the labels of all the reverse reaction templates along the unsolved reaction route are set to 0;

[0026] The empirical data set is obtained by searching the target compound library for empirical data collection.

[0027] As a preferred solution of a retrosynthetic analysis and processing device based on search experience, in the empirical model construction and training module, the empirical model is obtained by constructing a fully connected binary classification neural network model, and the empirical model is trained by the empirical data set and the binary cross entropy loss function; the vectorized representation data of the product and the vectorized representation data of the inverse reaction template are respectively input into the empirical model, processed by the empirical model, and the probability that the product can be solved by the reaction route after being processed by the inverse reaction template is output.

[0028] As a preferred solution of a retrosynthetic analysis and processing device based on search experience, in the pre-reactant generation module, in the expansion stage of the route search algorithm, the probability distribution of each reverse reaction template in the reverse reaction template library output by the SoftMax neural network model is used as the prior probability distribution of the reverse reaction template.

[0029] The invention has the following advantages: constructing a reverse reaction template library, the reverse reaction template library comprising a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field; constructing a SoftMax neural network model, and training the SoftMax neural network model through a cross entropy loss function and the reverse reaction template library; combining the trained SoftMax neural network model with a tree search algorithm to generate a route search algorithm; performing an iterative route search through the route search algorithm, setting labels for all reverse reaction templates along the reaction route, and obtaining an empirical data set; the empirical data set comprises a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field; constructing an empirical model, and training the empirical model through the empirical data set and a binary cross entropy loss function; performing a probability distribution adjustment based on the Bayesian theorem on the output probability of the empirical model, and obtaining a posterior probability distribution of the reverse reaction template; and generating a pre-reactant for the product through the route search algorithm according to the posterior probability distribution of the reverse reaction template. The present invention uses an empirical model trained by empirical data collected during the route planning process. The empirical model learns multi-step synthesizable information, which complements the single-step reaction information learned by the single-step model. The combination of the two can greatly improve the algorithm's solving ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0031] Figure 1 A schematic flow chart of a retrosynthetic analysis method based on search experience provided in Example 1 of the present invention;

[0032] Figure 2 A schematic diagram of probability distribution of an empirical model adjustment in an MCTS expansion phase in a retrosynthetic analysis method based on search experience provided in Example 1 of the present invention;

[0033] Figure 3 This is an architecture diagram of a reverse synthesis analysis processing device based on search experience provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0034] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] Example 1

[0036] See also Figure 1 Embodiment 1 of the present invention provides a retrosynthetic analysis method based on search experience, comprising the following steps:

[0037] S1. Construct a reverse reaction template library, wherein the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field;

[0038] S2, constructing a SoftMax neural network model, and training the SoftMax neural network model through a cross entropy loss function and the inverse reaction template library;

[0039] S3, combining the trained SoftMax neural network model with the tree search algorithm to generate a route search algorithm; performing iterative route search through the route search algorithm, setting labels for all reverse reaction templates along the reaction route, and obtaining an empirical data set; the empirical data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field;

[0040] S4, constructing an empirical model, and training the empirical model using the empirical data set and a binary cross entropy loss function;

[0041] S5. Adjusting the output probability of the empirical model according to the probability distribution of Bayes' theorem to obtain the posterior probability distribution of the reverse reaction template; and generating a pre-reactant for the product through the route search algorithm according to the posterior probability distribution of the reverse reaction template.

[0042] In this embodiment, in step S2, the SoftMax neural network model is obtained by constructing a fully connected multi-classification neural network model, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the vectorized representation data of the product is input into the SoftMax neural network model, processed by the SoftMax neural network model, and the probability distribution of each reverse reaction template in the reverse reaction template library is output.

[0043] Specifically, the vectorized representation of the product is a Morgan fingerprint with a length of 2048 and a radius of 2 calculated by RDKit; the multi-classification neural network is built using the Keras deep learning framework, the number of neurons in the input layer of the neural network is 2048, and the number of neurons in the output layer is N, where N is the number of non-repeated reverse reaction templates in the reverse reaction template library.

[0044] In this embodiment, in step S3, during the generation of the empirical data set, a target compound library is constructed, and the target compound library includes a compound SMILES field;

[0045] After the iterative route search is completed by the route search algorithm, the labels of all the reverse reaction templates along the solved reaction route are set to 1, and the labels of all the reverse reaction templates along the unsolved reaction route are set to 0;

[0046] The empirical data set is obtained by searching the target compound library for empirical data collection.

[0047] Specifically, Figure 2 As shown, the route search algorithm adopts the MCTS algorithm; the route search algorithm includes a selection phase, an expansion phase, and a backtracking phase; the number of algorithm iterations is set to 500, and the search tree depth is set to 6.

[0048] In this embodiment, in step S4, the empirical model is obtained by constructing a fully connected binary classification neural network model, and the empirical model is trained by the empirical data set and the binary cross entropy loss function; the vectorized representation data of the product and the vectorized representation data of the reverse reaction template are respectively input into the empirical model, processed by the empirical model, and the probability that the product can be solved by the reaction route after being processed by the reverse reaction template is output.

[0049] Specifically, the vectorized representation of the product and the vectorized representation of the reverse reaction template are both Morgan fingerprints with a length of 2048 and a radius of 2 calculated by RDKit; the Keras deep learning framework is used to build the binary classification neural network, and the neural network input layer is divided into two parts, which are used to receive the input of the product fingerprint and the reverse reaction template fingerprint respectively. The number of neurons is 2048, and the number of neurons in the output layer is 1.

[0050] In this embodiment, in step S5, during the expansion phase of the route search algorithm, the probability distribution of each reverse reaction template in the reverse reaction template library output by the SoftMax neural network model is used as the prior probability distribution of the reverse reaction template.

[0051] Specifically, the Bayesian theorem formula is simplified, and the posterior probability distribution calculation formula of the reverse reaction template is:

[0052] P 后验 =P 经验 ×P 先验

[0053] Where P 先验 is the probability distribution of the output of the SoftMax neural network model; P 经验 is the probability distribution of the empirical model output.

[0054] In summary, a reverse reaction template library is constructed, and the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field; a SoftMax neural network model is constructed, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the trained SoftMax neural network model is combined with a tree search algorithm to generate a route search algorithm; an iterative route search is performed by the route search algorithm, and labels are set for all reverse reaction templates along the reaction route to obtain an empirical data set; the empirical data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field; an empirical model is constructed, and the empirical model is trained by the empirical data set and a binary cross entropy loss function; the output probability of the empirical model is adjusted by a probability distribution based on the Bayesian theorem to obtain a posterior probability distribution of the reverse reaction template; and a pre-reactant is generated for the product by the route search algorithm according to the posterior probability distribution of the reverse reaction template. The present invention uses an empirical model trained by empirical data collected during the route planning process. The empirical model learns multi-step synthesizable information, which complements the single-step reaction information learned by the single-step model. The combination of the two can greatly improve the algorithm's solving ability.

[0055] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0056] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0057] Example 2

[0058] See also Figure 3 Embodiment 2 of the present invention further provides a retrosynthetic analysis processing device based on search experience, which adopts a retrosynthetic analysis method based on search experience of the above embodiment 1, comprising:

[0059] A reverse reaction template library construction module 001 is used to construct a reverse reaction template library, wherein the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field;

[0060] SoftMax neural network model construction and training module 002, used to construct a SoftMax neural network model, and train the SoftMax neural network model through a cross entropy loss function and the inverse reaction template library;

[0061] The empirical data set acquisition module 003 is used to combine the trained SoftMax neural network model with the tree search algorithm to generate a route search algorithm; perform iterative route search through the route search algorithm, set labels for all reverse reaction templates along the reaction route, and obtain an empirical data set; the empirical data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field;

[0062] An empirical model construction and training module 004 is used to construct an empirical model and train the empirical model using the empirical data set and a binary cross entropy loss function;

[0063] The pre-reactant generation module 005 is used to adjust the output probability of the empirical model based on the probability distribution of Bayes' theorem to obtain the posterior probability distribution of the reverse reaction template; and generate the pre-reactant for the product through the route search algorithm according to the posterior probability distribution of the reverse reaction template.

[0064] In this embodiment, in the SoftMax neural network model construction and training module 002, the SoftMax neural network model is obtained by constructing a fully connected multi-classification neural network model, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the vectorized representation data of the product is input into the SoftMax neural network model, processed by the SoftMax neural network model, and the probability distribution of each reverse reaction template in the reverse reaction template library is output.

[0065] In this embodiment, in the experience data set acquisition module 003, during the generation of the experience data set, a target compound library is constructed, and the target compound library includes a compound SMILES field;

[0066] After the iterative route search is completed by the route search algorithm, the labels of all the reverse reaction templates along the solved reaction route are set to 1, and the labels of all the reverse reaction templates along the unsolved reaction route are set to 0;

[0067] The empirical data set is obtained by searching the target compound library for empirical data collection.

[0068] In this embodiment, in the empirical model construction and training module 004, the empirical model is obtained by constructing a fully connected binary classification neural network model, and the empirical model is trained by the empirical data set and the binary cross entropy loss function; the vectorized representation data of the product and the vectorized representation data of the reverse reaction template are respectively input into the empirical model, processed by the empirical model, and the probability that the product can be solved by the reaction route after being processed by the reverse reaction template is output.

[0069] In this embodiment, in the pre-reactant generation module 005, in the expansion stage of the route search algorithm, the probability distribution of each reverse reaction template in the reverse reaction template library output by the SoftMax neural network model is used as the prior probability distribution of the reverse reaction template.

[0070] It should be noted that the information interaction, execution process and other contents between the modules of the above-mentioned device are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown in the previous part of the present application, and will not be repeated here.

[0071] Example 3

[0072] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code for a retrosynthetic analysis method based on search experience is stored, and the program code includes instructions for executing the retrosynthetic analysis method based on search experience of embodiment 1 or any possible implementation thereof.

[0073] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0074] Example 4

[0075] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0076] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute a retrosynthetic analysis method based on search experience of Example 1 or any possible implementation thereof.

[0077] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor implemented by reading software codes stored in a memory. The memory can be integrated in the processor or can be located outside the processor and exist independently.

[0078] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0079] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0080] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A retrosynthetic analysis method based on search experience, characterized in that: include: Constructing a reverse reaction template library, wherein the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field; Constructing a SoftMax neural network model, and training the SoftMax neural network model through a cross entropy loss function and the inverse reaction template library; The trained SoftMax neural network model is combined with the MCTS algorithm to generate a route search algorithm; an iterative route search is performed by the route search algorithm, and labels are set for all reverse reaction templates along the reaction route to obtain an empirical data set; the empirical data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field; Constructing an empirical model, and training the empirical model using the empirical data set and a binary cross entropy loss function; The output probability of the empirical model is adjusted for probability distribution based on Bayes' theorem to obtain the posterior probability distribution of the reverse reaction template; and the product is generated as a pre-reactant through the route search algorithm according to the posterior probability distribution of the reverse reaction template.

2. A retrosynthetic analysis method based on search experience according to claim 1, characterized in that: The SoftMax neural network model is obtained by constructing a fully connected multi-classification neural network model, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the vectorized representation data of the product is input into the SoftMax neural network model, processed by the SoftMax neural network model, and the probability distribution of each reverse reaction template in the reverse reaction template library is output.

3. A retrosynthetic analysis method based on search experience according to claim 2, characterized in that, During the generation of the empirical data set, a target compound library is constructed, wherein the target compound library includes a compound SMILES field; After the iterative route search is completed by the route search algorithm, the labels of all the reverse reaction templates along the solved reaction route are set to 1, and the labels of all the reverse reaction templates along the unsolved reaction route are set to 0; The empirical data set is obtained by searching the target compound library for empirical data collection.

4. A retrosynthetic analysis method based on search experience according to claim 3, characterized in that: The empirical model is obtained by constructing a fully connected binary classification neural network model, and the empirical model is trained by the empirical data set and the binary cross entropy loss function; the vectorized representation data of the product and the vectorized representation data of the reverse reaction template are respectively input into the empirical model, processed by the empirical model, and the probability that the product can be solved by the reaction route after being processed by the reverse reaction template is output.

5. A retrosynthetic analysis method based on search experience according to claim 4, characterized in that: In the expansion stage of the route search algorithm, the probability distribution of each reverse reaction template in the reverse reaction template library output by the SoftMax neural network model is used as the prior probability distribution of the reverse reaction template.

6. A retrosynthetic analysis processing device based on search experience, using a retrosynthetic analysis method based on search experience according to any one of claims 1 to 5, characterized in that: include: A reverse reaction template library construction module, used to construct a reverse reaction template library, wherein the reverse reaction template library includes a reactant SMILES field, a product SMILES field, and a reverse reaction template SMARTS field; A SoftMax neural network model construction and training module is used to construct a SoftMax neural network model and train the SoftMax neural network model through a cross entropy loss function and the inverse reaction template library; An experience data set acquisition module is used to combine the trained SoftMax neural network model with the MCTS algorithm to generate a route search algorithm; perform iterative route search through the route search algorithm, set labels for all reverse reaction templates along the reaction route, and obtain an experience data set; the experience data set includes a product SMILES field, a reverse reaction template SMARTS field, and a 0 / 1 label field; An empirical model construction and training module, used to construct an empirical model and train the empirical model using the empirical data set and a binary cross entropy loss function; The pre-reactant generation module is used to adjust the output probability of the empirical model based on the probability distribution of Bayes' theorem to obtain the posterior probability distribution of the reverse reaction template; and generate the pre-reactant for the product through the route search algorithm according to the posterior probability distribution of the reverse reaction template.

7. The retrosynthetic analysis processing device based on search experience according to claim 6, characterized in that: In the SoftMax neural network model construction and training module, the SoftMax neural network model is obtained by constructing a fully connected multi-classification neural network model, and the SoftMax neural network model is trained by a cross entropy loss function and the reverse reaction template library; the vectorized representation data of the product is input into the SoftMax neural network model, processed by the SoftMax neural network model, and the probability distribution of each reverse reaction template in the reverse reaction template library is output.

8. The retrosynthetic analysis processing device based on search experience according to claim 7, characterized in that: In the experience data set acquisition module, during the generation of the experience data set, a target compound library is constructed, wherein the target compound library includes a compound SMILES field; After the iterative route search is completed by the route search algorithm, the labels of all the reverse reaction templates along the solved reaction route are set to 1, and the labels of all the reverse reaction templates along the unsolved reaction route are set to 0; The empirical data set is obtained by searching the target compound library for empirical data collection.

9. The retrosynthetic analysis processing device based on search experience according to claim 8, characterized in that: In the empirical model construction and training module, the empirical model is obtained by constructing a fully connected binary classification neural network model, and the empirical model is trained using the empirical data set and a binary cross entropy loss function; the vectorized representation data of the product and the vectorized representation data of the reverse reaction template are respectively input into the empirical model, processed by the empirical model, and the probability that the product can be solved by the reaction route after being processed by the reverse reaction template is output.

10. The retrosynthetic analysis processing device based on search experience according to claim 9, characterized in that: In the pre-reactant generation module, in the expansion stage of the route search algorithm, the probability distribution of each reverse reaction template in the reverse reaction template library output by the SoftMax neural network model is used as the prior probability distribution of the reverse reaction template.

Images