Construction of drug intermediate database and AI intelligent retrieval method

Abstract

The present invention relates to the field of drug research and development technology, and discloses a drug intermediate database construction and AI intelligent retrieval method. The chemical structure and synthesis path data of the target drug intermediate are obtained, and the molecular descriptors are extracted using a preset feature extraction model, and a structured intermediate database is constructed accordingly. The database is then input into an AI retrieval model based on a graph neural network algorithm to generate dynamic retrieval results, and a preset optimization algorithm is used for multi-dimensional sorting, and a standardized retrieval list is output. The method also includes the steps of feature extraction model training, data preprocessing, retrieval result verification, and data security control. The present invention can efficiently integrate multi-source data, accurately extract key information, realize intelligent retrieval, and ensure data security, which can significantly improve drug research and development efficiency, reduce research and development costs, and provide strong support for drug research and development.

Description

Technical Field

[0001] The present invention relates to the technical field of drug research and development, and specifically to the construction of a drug intermediate database and an AI intelligent retrieval method. Background Art

[0002] In the field of drug research and development, drug intermediates serve as key raw materials for synthesizing final drugs, and the management and utilization of their related data are crucial to the R&D process. However, current drug intermediate data management and retrieval face many difficulties.

[0003] From a data acquisition perspective, chemical structure data and synthesis pathway data come from scattered sources. Chemical structure data is widely distributed across various chemical databases, such as SciFinder and Reaxys. Data formats and storage standards vary significantly across databases. Data generated by experimental recording equipment is also difficult to collect uniformly due to differences in equipment models and manufacturers. For example, the chemical structure data format recorded by a nuclear magnetic resonance instrument used in one laboratory is incompatible with that of another laboratory. This makes it difficult for researchers to integrate multi-source data, consuming significant time and effort on data format conversion and cleaning, severely impacting R&D efficiency.

[0004] When it comes to data processing, traditional methods struggle to effectively extract and utilize key information. The chemical structures of drug intermediates are complex, encompassing key features such as atomic connectivity, electron cloud distribution, and functional group topological relationships. However, existing technologies lack the means to efficiently and accurately extract these molecular descriptors. Researchers often rely on manual analysis of chemical structure data, a method that is not only inefficient but also prone to human error, leading to misjudgments of the properties and reactivity of drug intermediates.

[0005] In terms of database construction, most existing databases lack structured and multi-dimensional indexing. Data storage is fragmented and poorly correlated, making them unable to meet the needs of fast and accurate retrieval. When researchers need to find drug intermediates with specific structures and meeting certain synthesis criteria, they struggle to quickly locate relevant information in existing databases, hindering their R&D progress.

[0006] Search technology also has significant shortcomings. Existing search methods are mostly based on simple text matching or structural similarity searches, which fail to fully consider the various factors in drug development, such as reaction conditions, yield, synthesis difficulty, and cost. In actual R&D, researchers hope to find drug intermediates that meet structural requirements, have suitable synthesis conditions, and are relatively low in cost. However, traditional search methods cannot effectively provide this comprehensive information.

[0007] As competition in drug research and development becomes increasingly fierce, the requirements for R&D efficiency and innovation capabilities continue to increase. There is an urgent need for a method that can integrate multi-source data, efficiently extract key information, build a structured database and realize intelligent retrieval to solve the difficulties in existing technologies, accelerate the drug research and development process and reduce R&D costs. Summary of the Invention

[0008] The purpose of the present invention is to provide a pharmaceutical intermediate database construction and AI intelligent retrieval method to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solutions: a pharmaceutical intermediate database construction and AI intelligent retrieval method, the method comprising:

[0010] Obtaining chemical structure data and synthesis pathway data of a target drug intermediate, wherein the chemical structure data is derived from at least one chemical database or experimental recording device, and the synthesis pathway data includes reaction conditions, catalysts, and yield parameters;

[0011] Inputting the chemical structure data into a preset feature extraction model to identify and extract molecular descriptors from the chemical structure data, wherein the molecular descriptors include atomic connectivity, electron cloud distribution, and functional group topological relationships;

[0012] Constructing a structured intermediate database based on the molecular descriptors, wherein the database is associated with multi-dimensional index labels of chemical structure, synthesis pathway and physicochemical properties;

[0013] Inputting the structured intermediate database into a preset AI retrieval model to generate dynamic retrieval results, wherein the AI ​​retrieval model matches the similarity between user queries and database records based on a graph neural network algorithm;

[0014] The dynamic search results are sorted in multiple dimensions using a preset optimization algorithm, and a standardized search list is output.

[0015] Preferably, the step of training the feature extraction model includes: obtaining an initial training sample set, wherein each sample in the initial training sample set is labeled with a corresponding molecular descriptor label; iteratively training an initial feature extraction network using the initial training sample set until the matching rate between the feature prediction result output by the initial feature extraction network and the molecular descriptor label is greater than or equal to a preset threshold; during each iteration, adjusting the weight parameters of the initial feature extraction network according to the prediction error of the previous iteration, and eliminating samples in the initial training sample set with an error higher than the set threshold to form a reduced training subset for the next iterative training.

[0016] Preferably, obtaining the chemical structure data of the target drug intermediate includes obtaining chemical data from one of the data sources in the following manner:

[0017] Connect to the target chemical database server and / or real-time data interface of experimental equipment;

[0018] Extracting raw chemical data from at least one of the target chemical database server and the experimental equipment real-time data interface according to a preset data acquisition protocol, wherein the preset data acquisition protocol includes data format, storage accuracy and metadata association rules.

[0019] Preferably, before inputting the chemical structure data into a preset feature extraction model, the method further comprises:

[0020] filtering redundant information from the raw chemical data to eliminate format differences between different data sources;

[0021] A multi-scale noise reduction algorithm is used to remove outliers in the raw chemical data and retain key information of the molecular structure.

[0022] Preferably, constructing a structured intermediate database based on the molecular descriptors comprises:

[0023] segmenting the chemical structure map based on atomic connectivity in the molecular descriptors;

[0024] Based on the segmented graph, hierarchical storage units are generated, and the corresponding synthesis path and physicochemical property parameters are associated within each unit.

[0025] Preferably, inputting the structured intermediate database into a preset AI retrieval model comprises:

[0026] Setting query constraints in the AI ​​retrieval model to simulate user search requirements based on structural similarity, reaction conditions, or yield range;

[0027] By iteratively computing graph node embedding vectors, we can match database records with multi-dimensional similarity scores of query conditions.

[0028] Preferably, using a preset optimization algorithm to perform multi-dimensional sorting on the dynamic search results includes:

[0029] assigning weights to the dynamic search results to generate a derivative ranking list based on structural similarity, synthesis difficulty, or cost priority;

[0030] Based on the adaptive adjustment algorithm, the optimized parameters that conform to chemical logic are injected into the derived sorted list.

[0031] Preferably, the method further comprises training the optimization algorithm in the following manner:

[0032] Construct an optimized training set containing real chemical synthesis cases, where each case is annotated with the corresponding true value of the synthesis pathway;

[0033] The optimization algorithm is trained using a reinforcement learning framework so that the distribution difference between the ranking list generated by the optimization algorithm and the actual chemical synthesis case is less than a set tolerance range.

[0034] Preferably, the method further comprises:

[0035] Inputting the standardized search list into a preset verification model, and calculating the error matrix between the search results and the experimental record data;

[0036] Inversely adjust the graph network depth and feature embedding dimension of the AI ​​retrieval model according to the error matrix;

[0037] The verification model is constructed through the following steps:

[0038] Collect synthetic pathway data recorded in the laboratory as a verification benchmark;

[0039] An attention-based error evaluation network is established to output an error weight vector by comparing the retrieval results with the multi-dimensional feature graph of the verification benchmark.

[0040] Preferably, the method further comprises a data security control step:

[0041] Upon receiving an access request to the structured intermediate database from an external system, verifying that the identity token provided by the requester matches a preset permission list;

[0042] When the match is successful, the data subset that meets the access rights is transmitted to the external system via an encrypted link.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The pharmaceutical intermediate database construction and AI intelligent retrieval method proposed in this invention have brought significant benefits in many aspects to the field of drug research and development. In terms of data acquisition and processing, this method can obtain chemical structure data and synthesis pathway data from a variety of chemical databases and experimental recording equipment. By presetting the data acquisition protocol, the data format, storage accuracy, and metadata association rules of different data sources can be unified, greatly improving the efficiency and accuracy of data acquisition. At the same time, before the data is input into the feature extraction model, redundant information filtering and multi-scale noise reduction processing will be performed to effectively eliminate format differences and outliers, retain key information, provide a high-quality data foundation for subsequent analysis, and avoid research bias caused by data quality issues.

[0045] The feature extraction model has been meticulously trained to accurately identify and extract molecular descriptors from chemical structure data, such as atomic connectivity, electron cloud distribution, and functional group topology. This process not only improves the efficiency of extracting key information but also provides an important basis for subsequent database construction and intelligent retrieval. Through iterative training and sample screening on a large number of samples, the model's accuracy and stability have been continuously improved. Compared to traditional manual analysis, this significantly reduces human error and enables researchers to gain a deeper and more accurate understanding of the chemical properties of drug intermediates.

[0046] The construction of a structured intermediate database is a highlight. It segments chemical structure maps based on molecular descriptors and generates hierarchical storage units, while also associating multi-dimensional index tags with chemical structure, synthesis pathway, and physicochemical properties. This structured design significantly improves data storage and retrieval efficiency. Researchers can quickly locate required data from multiple dimensions, such as screening drug intermediates based on specific atomic connections, reaction conditions, or physicochemical properties. This significantly reduces data search time and improves the overall efficiency of R&D.

[0047] The AI ​​retrieval model, based on a graph neural network algorithm, can more accurately match the similarity between user queries and database records, generating dynamic search results. Compared to traditional search methods, this model comprehensively considers multiple factors, such as structural similarity, reaction conditions, and yield, providing users with search results that are more tailored to their actual needs. Furthermore, a preset optimization algorithm is used to sort search results across multiple dimensions, outputting a standardized search list. Through weight assignment and adaptive adjustment algorithms, researchers can prioritize search results based on their needs, such as focusing on structural similarity, synthesis difficulty, or cost, further improving the practicality and pertinence of searches.

[0048] Furthermore, the present invention incorporates a data security control step to verify the matching of the requester's identity token with a preset permission list and transmit data via an encrypted link. This effectively safeguards the security of sensitive data in the database, prevents data leaks, meets the stringent data security requirements of the pharmaceutical R&D industry, and provides reliable protection for the data assets of enterprises and research institutions. Overall, this invention comprehensively improves the management and utilization of pharmaceutical intermediate data, significantly advances drug R&D progress, reduces R&D costs, and has significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a working principle diagram of the pharmaceutical intermediate database construction and AI intelligent retrieval method of the present invention;

[0050] Figure 2 Flowchart for feature extraction model training;

[0051] Figure 3 Flowchart for chemical structure data preprocessing;

[0052] Figure 4 Flowchart for multi-dimensional sorting of search results. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.

[0054] See also Figures 1-4 The present invention provides a technical solution: The present invention relates to a pharmaceutical intermediate database construction and AI intelligent retrieval method, and the specific implementation steps are as follows:

[0055] Data Acquisition: Obtain chemical structure data and synthesis pathway data for the target drug intermediate. Chemical structure data can be obtained from a wide range of sources, including at least one chemical database, such as the commonly used ChemSpider and PubChem, which store chemical structure information for a large number of known compounds. It can also be derived from experimental recording equipment, such as experimental data recorded by high-resolution spectrometers, chromatographs, and other equipment during laboratory drug intermediate synthesis experiments. Synthesis pathway data includes reaction conditions, such as reaction temperature, reaction time, and reaction pressure; catalysts (substances that participate in chemical reactions but whose chemical properties remain unchanged before and after the reaction); and yield parameters, which are used to measure the efficiency of the synthesis reaction in producing the target drug intermediate.

[0056] Extracting molecular descriptors: The acquired chemical structure data is fed into a pre-set feature extraction model. This model identifies and extracts molecular descriptors from the chemical structure data. Molecular descriptors include atomic connectivity, which describes the connection and topological structure between atoms in a molecule; electron cloud distribution, which reflects the distribution of electrons in a molecule and significantly influences its chemical activity and reactivity; and functional group topology, which defines the spatial location and interactions between different functional groups in a molecule.

[0057] Constructing a structured intermediate database: Based on the extracted molecular descriptors, a structured intermediate database is constructed. This database associates multi-dimensional index tags with chemical structure, synthesis pathway, and physicochemical properties, enabling rapid retrieval and analysis of data from multiple perspectives, making it easier for researchers to obtain the information they need.

[0058] Generate dynamic search results: Input the constructed structured intermediate database into a preset AI search model. This AI search model is based on a graph neural network algorithm. By matching the similarity between the user query and the database records, it generates dynamic search results and provides users with pharmaceutical intermediate information related to the query.

[0059] Multi-dimensional sorting and output: Dynamic search results are sorted across multiple dimensions using a pre-set optimization algorithm. Taking into account various factors such as structural similarity, synthesis difficulty, and cost, the sorted results are output as a standardized search list for easy viewing and use.

[0060] The present invention will be further described below in conjunction with Examples 1 to 5:

[0061] Example 1:

[0062] In this embodiment, the specific process of training the feature extraction model is mainly described.

[0063] Obtain an initial training sample set, in which each sample is annotated with a corresponding molecular descriptor label. These samples can be collected from public chemical datasets or large amounts of experimental data. For example, select data containing the chemical structures of drug intermediates and their corresponding molecular descriptors from compound datasets published by professional chemical research institutions as the initial sample.

[0064] The initial feature extraction network is iteratively trained using the initial training sample set. During the training process, the network weight parameters are continuously adjusted to make the feature prediction results output by the network match the molecular descriptor labels as closely as possible. The concept of matching rate is introduced here. After each training is completed, the matching rate between the prediction results and the labels is calculated. Let the prediction result be , the molecular descriptor label is , matching rate The calculation formula is:

[0065]

[0066] in is the sample size, is the judgment function, when and When matching, , otherwise When the matching rate is greater than or equal to the preset threshold, the training ends.

[0067] During each iteration, the weight parameters of the initial feature extraction network are adjusted according to the prediction error of the previous iteration. The prediction error can be calculated by the mean square error (MSE), assuming that the prediction value is , the true value is , then the mean square error

[0068]

[0069] in is the number of samples in the current training batch. At the same time, samples with errors above a set threshold in the initial training set are removed to form a reduced training subset for the next iteration of training. This gradually optimizes the network and improves the accuracy of feature extraction.

[0070] Example 2:

[0071] This example describes in detail a specific method for obtaining the chemical structure data of a target drug intermediate.

[0072] To acquire chemical structure data, you can connect to a target chemical database server and / or a real-time data interface for experimental equipment. For example, connecting to the commonly used chemical database ChemSpider uses its API interface and programming code written in a specific programming language (such as Python) to achieve communication with the database server. For a real-time data interface for experimental equipment, suppose your laboratory uses a certain model of nuclear magnetic resonance spectrometer equipped with a dedicated data transmission interface. Using the corresponding data transmission line and driver, the data collected by the instrument is transmitted to a computer in real time.

[0073] Raw chemical data is extracted from at least one of the target chemical database server and the real-time data interface of the experimental equipment according to a preset data acquisition protocol. The preset data acquisition protocol includes data format, storage precision, and metadata association rules. In terms of data format, for example, data extracted from the database may be in JSON format, and data obtained from the experimental equipment may be in CSV format. Storage precision is set according to actual needs. For example, for chemical shift data, the storage precision can be set to three decimal places. Metadata association rules are used to clarify the relationship between data and other related information, such as the relationship between the name, source, and other information of the compound and chemical structure data.

[0074] Before inputting chemical structure data into the pre-set feature extraction model, the raw chemical data must be processed. Redundant information is first filtered out. Data from different data sources may contain duplicate or irrelevant information. By writing specific program code, this redundant information is removed based on the data's characteristics and patterns. This also eliminates format differences between different data sources, unifying all data into a format suitable for subsequent processing. A multiscale noise reduction algorithm is then used to remove outliers from the raw chemical data. Based on the principles of wavelet transforms, this algorithm decomposes the data into different scale spaces, estimates and removes noise at each scale, retains key molecular structure information, improves data quality, and provides a reliable data foundation for subsequent feature extraction.

[0075] Example 3:

[0076] When constructing a structured intermediate database, we use a specific pharmaceutical intermediate, ibuprofen, as an example. The chemical structure of ibuprofen contains multiple complex atomic connections and functional groups, which are interconnected to form a specific chemical structure map.

[0077] The chemical structure map of an ibuprofen intermediate was segmented based on the atomic connectivity in its molecular descriptors. First, the atomic connectivity data was analyzed to determine the bonding relationships and topological structure between atoms in the map. For example, an analysis of its chemical structure revealed that the carbon atoms on the benzene ring form stable covalent bonds with other atoms and have a specific connection order and spatial relationship with atoms on the side chains. Graph theory algorithms, such as the minimum spanning tree (MST), were used to construct a tree structure with the minimum sum of edge weights while ensuring all atoms are connected. Starting from the chemical structure map of the ibuprofen intermediate, edges were gradually added based on atomic connectivity to construct a minimum spanning tree. This process partitioned the map into several sub-graphs, each representing a relatively independent and functionally specific portion of the chemical structure. For example, the benzene ring could be considered as one sub-graph, while the carboxyl group and associated atoms on the side chain could be considered as another sub-graph.

[0078] Hierarchical storage units are generated based on the segmented graph. These subgraphs are used as nodes in the hierarchical structure to construct a tree-like storage system. A hierarchical relationship is established, with the node corresponding to the benzene ring subgraph as the parent node and the closely related side chain subgraph nodes as child nodes. Within each storage unit, the corresponding synthesis pathway and physicochemical property parameters are associated. For a storage unit for the benzene ring portion of an ibuprofen intermediate, the synthesis pathway parameters might include: In a reaction, benzene as the starting material undergoes a Friedel-Crafts alkylation reaction with a halogenated hydrocarbon at a specific temperature (e.g., 50-60°C) and pressure (1-2 atm) using aluminum trichloride as a catalyst, with a reaction time of 3-4 hours. Physicochemical properties include a melting point of [X]°C and a solubility of [X]g / 100mL in common organic solvents such as ethanol. For a storage unit related to a side chain carboxyl group, the synthesis pathway might involve a specific oxidation reaction using a specific oxidant (e.g., potassium permanganate) under alkaline conditions, with a temperature of [X]°C and a reaction time of [X] hours. Physicochemical properties, such as solubility in water and acid dissociation constant, are also stored within this unit. This structured storage of complex drug intermediate chemical structures, synthesis pathways, and physicochemical properties facilitates subsequent data query, analysis, and management.

[0079] Example 4:

[0080] When inputting the structured intermediate database into the preset AI search model, let's take the search related to ibuprofen intermediates as an example. Suppose the user wants to find pharmaceutical intermediates with similar structures to ibuprofen intermediates and higher yields, and set query constraints in the AI ​​search model.

[0081] For the retrieval of structural similarity, the model uses a similarity calculation method based on molecular fingerprints. Molecular fingerprints are a way to encode molecular structures, which converts the structural information of molecules into a series of binary numbers. In this embodiment, the extended connectivity fingerprint (ECFP) is selected to describe the molecular structure. ECFP searches for substructures with increasing radius in the molecular structure and generates fingerprints by hashing these substructures. The structural similarity threshold is set to 0.7, that is, when the ECFP fingerprint similarity between the target molecule and the molecule in the database is greater than 0.7, the two are considered to have similar structures. The similarity calculation uses the Tanimoto coefficient, and its calculation formula is:

[0082]

[0083] in, is the number of bits that are 1 in both molecular fingerprints, and are the total number of 1 bits in the two molecular fingerprints respectively.

[0084] For a yield range search, suppose the user wants to find pharmaceutical intermediates with a yield greater than 75%. In the model, set the yield query condition to greater than 75%.

[0085] By iteratively calculating the graph node embedding vector, the multi-dimensional similarity score between the database record and the query condition is matched. The Graph Attention Network (GAT) algorithm is used to calculate the graph node embedding vector. In GAT, each node updates its own feature representation based on the features and relative importance of its neighboring nodes. For each drug intermediate molecule in the database, its chemical structure map is converted into graph structure data and input into the GAT model. Assume that the node In the The eigenvector at the iteration is , then the update formula is:

[0086]

[0087] in, is an activation function, such as the ReLU function; It is The weight matrix at the iteration; is a node The set of neighbor nodes of is a node and nodes The attention coefficient between is calculated by the following formula:

[0088]

[0089] here is a learnable parameter vector, Represents a vector concatenation operation. Through multiple iterations, a stable node embedding vector is obtained. Then, combined with structural similarity and yield conditions, a multidimensional similarity score is calculated between the database record and the query conditions. For each drug intermediate in the database, a multidimensional similarity score is calculated based on its structural similarity score (calculated based on the Tanimoto coefficient) and whether the yield meets the conditions. For example, the structural similarity score accounts for 0.6, and the yield score accounts for 0.4 (weights are set according to user requirements). Finally, a comprehensive score is obtained for each record, which is sorted by score. Search results that meet the user's query requirements are output, providing valuable reference information for drug developers.

[0090] Example 5:

[0091] This embodiment focuses on the use of a preset optimization algorithm to perform multi-dimensional sorting of dynamic retrieval results and the related training process, and also involves verification model related content. Weights are assigned to the dynamic retrieval results to generate a derivative sorting list based on structural similarity, synthesis difficulty or cost priority. For example, for structural similarity, a higher weight is given based on the similarity score calculated previously; for synthesis difficulty, a quantitative assessment can be made based on factors such as the number of steps required for the synthesis reaction and the severity of the reaction conditions, and corresponding weights are assigned; for cost, raw material cost, catalyst cost, equipment cost, etc. are considered, and weights are assigned after comprehensive evaluation. Assume that the structural similarity weight is , the synthetic difficulty weight is , the cost weight is ,and ,The comprehensive score of each search result is obtained through weighted calculation, thereby generating a derivative ranking list.

[0092] Based on the adaptive adjustment algorithm, the optimization parameters that conform to chemical logic are injected into the derivative ranking list. The adaptive adjustment algorithm can dynamically adjust the weight according to the feedback information of the search results. For example, if multiple searches find that users are more interested in results with high structural similarity and low synthesis difficulty, the algorithm can appropriately increase the weight of structural similarity. , reduce the weight of synthesis difficulty .

[0093] When training the optimization algorithm, an optimization training set containing real chemical synthesis cases is constructed, and each case is annotated with the corresponding true value of the synthesis path. A large number of real chemical synthesis cases are collected from actual chemical synthesis literature and experimental records, such as collecting 1,000 synthesis cases of different drug intermediates. The optimization algorithm is trained using a reinforcement learning framework so that the distribution difference between the sorted list generated by the optimization algorithm and the real chemical synthesis case is less than the set tolerance range. The distribution difference can be measured by calculating the Kullback-Leibler divergence (KL divergence). Let the distribution of the sorted list generated by the optimization algorithm be , the distribution of real chemical synthesis cases is , then KL divergence ,

[0094] in is the length of the sorted list. By continuously adjusting the parameters of the optimization algorithm, the KL divergence is made smaller than the set tolerance range, thereby improving the performance of the optimization algorithm.

[0095] The entire retrieval process also involves a verification model. The standardized retrieval list is input into the preset verification model, and the error matrix between the retrieval results and the experimental record data is calculated. The steps for constructing the verification model are: collecting the synthesis path data recorded in the laboratory as a verification benchmark, and extracting a large amount of reliable synthesis path data from the laboratory's experimental record database. An error evaluation network based on an attention mechanism is established, and an error weight vector is output by comparing the multi-dimensional feature graph of the retrieval results and the verification benchmark. The attention mechanism can make the network pay more attention to the differences between the important features in the retrieval results and the verification benchmark, improve the accuracy of error evaluation, and then reversely adjust the graph network depth and feature embedding dimension of the AI ​​retrieval model according to the error matrix, and continuously optimize the performance of the retrieval model. At the same time, the present invention also includes a data security control step. When receiving an access request from an external system to the structured intermediate database, the identity token provided by the requester is verified to match the preset permission list. If the match is successful, the data subset that meets the access rights is transmitted to the external system via an encrypted link to ensure the security and confidentiality of the data.

[0096] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0097] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A pharmaceutical intermediate database construction and AI intelligent retrieval method, characterized in that: include: Obtaining chemical structure data and synthesis pathway data of a target drug intermediate, wherein the chemical structure data is derived from at least one chemical database or experimental recording device, and the synthesis pathway data includes reaction conditions, catalysts, and yield parameters; Inputting the chemical structure data into a preset feature extraction model to identify and extract molecular descriptors from the chemical structure data, wherein the molecular descriptors include atomic connectivity, electron cloud distribution, and functional group topological relationships; Constructing a structured intermediate database based on the molecular descriptors, wherein the database is associated with multi-dimensional index labels of chemical structure, synthesis pathway and physicochemical properties; Inputting the structured intermediate database into a preset AI retrieval model to generate dynamic retrieval results, wherein the AI ​​retrieval model matches the similarity between user queries and database records based on a graph neural network algorithm; Using a preset optimization algorithm to perform multi-dimensional sorting on the dynamic search results and output a standardized search list; Constructing a structured intermediate database based on the molecular descriptors includes: segmenting the chemical structure map based on atomic connectivity in the molecular descriptors; Based on the segmented graph, hierarchical storage units are generated, and the corresponding synthesis path and physicochemical property parameters are associated within each unit.

2. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 1, characterized in that: The steps of training the feature extraction model include: obtaining an initial training sample set, wherein each sample in the initial training sample set is labeled with a corresponding molecular descriptor label; iteratively training an initial feature extraction network using the initial training sample set until the matching rate between the feature prediction results output by the initial feature extraction network and the molecular descriptor label is greater than or equal to a preset threshold; during each iteration, adjusting the weight parameters of the initial feature extraction network according to the prediction error of the previous iteration, and eliminating samples in the initial training sample set with errors higher than the set threshold to form a reduced training subset for the next iterative training.

3. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 1, characterized in that: Obtaining the chemical structure data of the target drug intermediate includes obtaining chemical data from one of the data sources in the following manner: Connect to the target chemical database server and / or real-time data interface of experimental equipment; Extracting raw chemical data from at least one of the target chemical database server and the experimental equipment real-time data interface according to a preset data acquisition protocol, wherein the preset data acquisition protocol includes data format, storage accuracy and metadata association rules.

4. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 3, characterized in that: Before inputting the chemical structure data into a preset feature extraction model, the method further includes: filtering redundant information from the raw chemical data to eliminate format differences between different data sources; A multi-scale noise reduction algorithm is used to remove outliers in the raw chemical data and retain key information of the molecular structure.

5. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 1, characterized in that: Inputting the structured intermediate database into a preset AI retrieval model comprises: Setting query constraints in the AI ​​retrieval model to simulate user search requirements based on structural similarity, reaction conditions, or yield range; By iteratively computing graph node embedding vectors, we can match database records with multi-dimensional similarity scores of query conditions.

6. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 1, characterized in that: Using a preset optimization algorithm to perform multi-dimensional sorting on the dynamic search results includes: assigning weights to the dynamic search results to generate a derivative ranking list based on structural similarity, synthesis difficulty, or cost priority; Based on the adaptive adjustment algorithm, the optimized parameters that conform to chemical logic are injected into the derived sorted list.

7. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 6, characterized in that: The method further comprises training the optimization algorithm as follows: Construct an optimized training set containing real chemical synthesis cases, where each case is annotated with the corresponding true value of the synthesis pathway; The optimization algorithm is trained using a reinforcement learning framework so that the distribution difference between the ranking list generated by the optimization algorithm and the actual chemical synthesis case is less than a set tolerance range.

8. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 1, characterized in that: The method further comprises: Inputting the standardized search list into a preset verification model, and calculating the error matrix between the search results and the experimental record data; Inversely adjust the graph network depth and feature embedding dimension of the AI ​​retrieval model according to the error matrix; The verification model is constructed through the following steps: Collect synthetic pathway data recorded in the laboratory as a verification benchmark; An error evaluation network based on the attention mechanism is established, which outputs the error weight vector by comparing the retrieval results with the multi-dimensional feature graph of the verification benchmark.

9. The pharmaceutical intermediate database construction and AI intelligent retrieval method according to claim 1, characterized in that: The method further comprises a data security control step: Upon receiving an access request to the structured intermediate database from an external system, verifying that the identity token provided by the requester matches a preset permission list; When the match is successful, the data subset that meets the access rights is transmitted to the external system via an encrypted link.

Images