Artificial intelligence-based drug molecule processing method, device, equipment and storage medium

Abstract

The application provides an artificial intelligence-based drug molecule processing method and device, electronic equipment and computer readable storage medium; relates to artificial intelligence technology; the method comprises the following steps: determining a plurality of candidate drug molecules for a target protein; performing activity prediction processing based on the plurality of candidate drug molecules and the target protein to obtain activity information of each candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each candidate drug molecule; and performing screening processing on the plurality of candidate drug molecules based on the activity information of each candidate drug molecule and the molecular docking information of each candidate drug molecule, and taking the screened candidate drug molecule as a target drug molecule. Through the application, valuable drug molecules can be efficiently screened from a large number of candidate drug molecules.

Description

Technical Field

[0001] This application relates to smart medical technology, and more particularly to a method, apparatus, electronic device, and computer-readable storage medium for processing drug molecules based on artificial intelligence. Background Technology

[0002] Artificial Intelligence (AI) is a comprehensive technology within computer science that studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. AI technology is a multidisciplinary field, encompassing a wide range of areas, including natural language processing and machine learning / deep learning. With technological advancements, AI will be applied in more fields and play an increasingly important role.

[0003] Drug screening is a technique that identifies target drug molecules from a large pool of drug molecules that may be valuable for certain diseases (such as polycystic ovary syndrome). Currently, compound libraries contain a large number of potentially valuable candidate drug molecules; however, extracting valuable data from these candidates relies on manual screening, which is neither accurate nor efficient, and is extremely costly.

[0004] There is a lack of solutions for efficiently screening candidate drug molecules based on artificial intelligence. Summary of the Invention

[0005] This application provides an artificial intelligence-based drug molecule processing method, apparatus, electronic device, and computer-readable storage medium, which enables efficient screening of valuable drug molecules from a large number of candidate drug molecules.

[0006] The technical solution of this application embodiment is implemented as follows:

[0007] This application provides an artificial intelligence-based drug molecule processing method, including:

[0008] Identify multiple candidate drug molecules targeting the target protein;

[0009] Based on the multiple candidate drug molecules and the target protein, activity prediction processing is performed to obtain the activity information of each candidate drug molecule;

[0010] Homology modeling is performed on the target protein to obtain a reference protein with a homologous structure to the target protein;

[0011] Molecular docking is performed based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information for each candidate drug molecule.

[0012] Based on the activity information and molecular docking information of each candidate drug molecule, the plurality of candidate drug molecules are screened, and the screened candidate drug molecules are used as target drug molecules.

[0013] This application provides an artificial intelligence-based drug molecule processing device, comprising:

[0014] The identification module is used to identify multiple candidate drug molecules targeting the target protein;

[0015] The prediction module is used to perform activity prediction processing based on the multiple candidate drug molecules and the target protein to obtain the activity information of each candidate drug molecule.

[0016] The processing module is used to perform homology modeling on the target protein to obtain a reference protein with a homology structure to the target protein; and to perform molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information for each candidate drug molecule.

[0017] The screening module is used to screen the multiple candidate drug molecules based on the activity information and molecular docking information of each candidate drug molecule, and to select the screened candidate drug molecules as target drug molecules.

[0018] In the above technical solution, the determining module is further used to screen compounds in the compound library based on the target protein to obtain multiple screened compounds;

[0019] The selected compounds are pretreated and used as candidate drug molecules targeting the target protein.

[0020] In the above technical solution, the determining module is also used to perform molecular drug-making rule-based screening on compounds in the compound library based on the target protein to obtain multiple compounds that meet the drug-making rules;

[0021] The compounds that meet the drug formulation rules are subjected to deduplication, and the deduplicated compounds are used as the screened compounds.

[0022] In the above technical solution, the determining module is further used to perform chemical filtration on the plurality of screened compounds based on the target group to obtain a plurality of filtered compounds;

[0023] The enantiomers of the chiral compounds are removed from the plurality of filtered compounds, and the compounds from which the enantiomers are removed are selected as candidate drug molecules for the target protein.

[0024] In the above technical solution, the prediction module is further configured to perform the following processing on any one of the plurality of candidate drug molecules:

[0025] The molecular structure of the candidate drug molecule is encoded to obtain the embedding features of the candidate drug molecule;

[0026] The protein structure of the target protein is encoded to obtain the embedding features of the target protein;

[0027] The embedding features of the candidate drug molecule and the embedding features of the target protein are fused to obtain an active fusion feature;

[0028] The activity fusion features are mapped to obtain the activity information of the candidate drug molecule.

[0029] In the above technical solution, the prediction module is further used to construct a molecular map of the candidate drug molecule based on the molecular structure of the candidate drug molecule;

[0030] The molecular map of the candidate drug molecule is image encoded to obtain the embedding features of the candidate drug molecule.

[0031] In the above technical solution, the prediction module is further used to determine the protein sequence of the target protein based on the protein structure of the target protein;

[0032] The protein sequence of the target protein is subjected to text conversion processing to obtain the embedding features of the target protein.

[0033] In the above technical solution, the prediction module is further configured to sum the embedding features of the candidate drug molecule and the embedding features of the target protein, and use the result of the summing process as the activity fusion feature; or,

[0034] The embedding features of the candidate drug molecule and the embedding features of the target protein are spliced ​​together, and the result of the splicing process is used as the active fusion feature.

[0035] In the above technical solution, the prediction module is further used to map the embedding features of the candidate drug molecule and the embedding features of the target protein to obtain an intermediate feature vector containing the candidate drug molecule and the target protein.

[0036] An affine transformation is performed on the intermediate feature vector to obtain the active fusion feature.

[0037] In the above technical solution, the prediction module is further used to map the active fusion feature to the latent vector space to obtain the latent vector of the active fusion feature;

[0038] The latent vector of the activity fusion feature is subjected to nonlinear mapping processing, and the resulting activity score of the candidate drug molecule is used as the activity information of the candidate drug molecule.

[0039] In the above technical solution, the processing module is further configured to perform the following processing on any candidate protein in the protein library:

[0040] The similarity between the candidate protein sequence and the target protein sequence is calculated by performing similarity analysis.

[0041] When the similarity is greater than the similarity threshold, structural optimization is performed based on the three-dimensional structure of the candidate protein, and the result of the structural optimization is used as a reference protein with a homologous structure to the target protein.

[0042] In the above technical solution, the processing module is further used to perform molecular dynamics simulation based on the reference protein to obtain the active site and binding pocket of the reference protein;

[0043] The candidate drug molecules are pretreated to obtain the molecular conformation of each candidate drug molecule;

[0044] For each candidate drug molecule, the following processing is performed: molecular docking scoring is performed based on the active site of the reference protein, the binding pocket, and the molecular conformation of the candidate drug molecule, and the results of the molecular docking scoring are used as the molecular docking information of the candidate drug molecule.

[0045] In the above technical solution, the processing module is further used to perform format conversion processing on the plurality of candidate drug molecules respectively to obtain the conversion format of each candidate drug molecule;

[0046] Based on the transformation format of each candidate drug molecule, a three-dimensional conformation of each candidate drug molecule is constructed;

[0047] Based on the three-dimensional conformation of each candidate drug molecule, the positions of hydrogen atoms that can be added are determined;

[0048] Add hydrogen atoms to the positions where hydrogen atoms can be added to obtain the molecular conformation of the candidate drug molecule.

[0049] In the above technical solution, the screening module is also used to perform clustering processing on the multiple candidate drug molecules to obtain multiple drug category sets;

[0050] From the multiple drug category sets, candidate drug molecules that meet the requirements for activity information and molecular docking information are selected as the target drug molecules.

[0051] In the above technical solution, the screening module is further configured to perform the following processing on any one of the multiple drug category sets:

[0052] The candidate drug molecule with the highest activity information in the set of drug categories is selected as the drug molecule to be screened.

[0053] The activity information, molecular docking information, and drug properties of the drug molecule to be screened are weighted and summed to obtain the comprehensive drug information of the drug molecule to be screened.

[0054] Based on the comprehensive drug information of multiple drug molecules to be screened, the multiple drug molecules to be screened are sorted in descending order, and the drug molecules to be screened at the top of the sort are taken as the target drug molecules.

[0055] This application provides an electronic device for drug molecule processing, the electronic device comprising:

[0056] Memory, used to store executable instructions;

[0057] The processor, when executing executable instructions stored in the memory, implements the artificial intelligence-based drug molecule processing method provided in the embodiments of this application.

[0058] This application provides a computer-readable storage medium storing executable instructions for inducing a processor to execute and implement the artificial intelligence-based drug molecule processing method provided in this application.

[0059] The embodiments of this application have the following beneficial effects:

[0060] Based on the activity and molecular docking information of candidate drug molecules, target drug molecules targeting the target protein are screened from a large number of candidate drug molecules, thereby ensuring the activity of the target drug molecules and their binding ability to the target protein, so as to efficiently and automatically screen valuable target drug molecules. Attached Figure Description

[0061] Figure 1 This is a schematic diagram illustrating the application scenario of the medical system provided in the embodiments of this application;

[0062] Figure 2 This is a schematic diagram of the structure of an electronic device for drug molecule processing provided in an embodiment of this application;

[0063] Figures 3-5 This is a flowchart illustrating the artificial intelligence-based organic sample processing method provided in the embodiments of this application;

[0064] Figure 6This is a schematic diagram of the structure of the prediction model provided in the embodiments of this application;

[0065] Figure 7 This is a schematic diagram of the G protein-coupled receptor GPR54 / Kiss-1 signaling pathway provided by related technologies;

[0066] Figure 8 This is a schematic diagram illustrating the structural optimization of lead compound 1 provided by related technologies;

[0067] Figure 9 This is a schematic diagram of the process for scoring compound activity based on deep learning, provided in an embodiment of this application.

[0068] Figures 10A-10B This is a schematic diagram of the structure of a potential active molecule provided in the embodiments of this application. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0070] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0072] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0073] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0074] 1) Deep Learning (DL): A new research direction in the field of Machine Learning (ML), introduced to bring machine learning closer to its original goal—Artificial Intelligence (AI). Deep learning learns the inherent patterns and hierarchical representations of sample data. The information gained during this learning process greatly helps in interpreting data such as text, images, and sound. The ultimate goal of deep learning is to enable machines to have analytical and learning capabilities like humans, and to recognize data such as text, images, and sound.

[0075] 2) Convolutional Neural Networks (CNNs): A class of feedforward neural networks (FNNs) that include convolutional computations and have a deep structure, CNNs are one of the representative algorithms of deep learning. CNNs possess representation learning capabilities, enabling them to perform shift-invariant classification of input images according to their hierarchical structure.

[0076] 3) Simplified Molecular Input Line Entry Specification (SMILES): A specification that explicitly describes molecular structures using American Standard Code for Information Interchange (ASCII) strings. A SMILES expression can describe a three-dimensional chemical structure using a string of characters. For example, the SMILES expression for cyclohexane (C6H12) is C1CCCCC1, meaning C1CCCCC1 represents cyclohexane.

[0077] 4) Drug molecule: The chemical structure of a drug, for example, a benzene ring is a drug molecule.

[0078] 5) Drug properties: Properties that characterize the structure of drug molecules, such as solubility, blood-brain barrier permeability, and toxicity (the ability of a certain type of drug molecule to cause direct or indirect damage after contact with a living organism or entering a living organism, i.e., the biological toxicity of that type of drug molecule).

[0079] 6) Molecular docking: A crucial method in molecular simulation, molecular docking is a drug design approach that utilizes receptor characteristics and the interaction mechanisms between receptors and drug molecules. Essentially, it involves the recognition process between multiple molecules, encompassing spatial and energy matching. In drug development, it's primarily used to calculate and visualize the binding patterns and key interactions between small molecules and target proteins. Molecular docking methods can be simplified to some extent, and based on the degree and method of simplification, they can be categorized into rigid docking, semi-flexible docking, and flexible docking.

[0080] 7) Homology Modeling: The three-dimensional structure of proteins is crucial for understanding their biological and physiological functions and for drug design based on target structures. Homology modeling is a method that starts with the amino acid sequence of a protein and uses the experimentally resolved three-dimensional structures of homologous proteins as templates to construct the three-dimensional structure of the target protein.

[0081] 8) Binding Pocket: In drug design, the cavity in which small molecules bind to proteins to regulate their functions is called a binding pocket.

[0082] 9) G protein-coupled receptors (GPCRs): This is a general term for a large class of membrane protein receptors. These receptors share the common feature of having seven transmembrane α-helices in their three-dimensional structure, and G protein binding sites on the C-terminus of their polypeptide chain and the intracellular loop connecting the 5th and 6th transmembrane helices.

[0083] 10) Lead compounds: Abbreviated as lead compounds, are compounds with certain biological activities and chemical structures obtained through various routes and means, which are used for further modification of compound structures.

[0084] This application provides an artificial intelligence-based drug molecule processing method, apparatus, electronic device, and computer-readable storage medium, which can efficiently screen valuable target drug molecules from a large number of candidate drug molecules.

[0085] The AI-based drug molecule processing method provided in this application can be implemented by a terminal / server alone; or it can be implemented collaboratively by a terminal and a server. For example, the terminal can independently undertake the AI-based drug molecule processing method described below, or the terminal can send a screening request for drug molecules (including target protein information) to the server. The server executes the AI-based drug molecule processing method according to the received screening request to obtain the target drug molecule for the target protein. Pharmaceutical researchers can then quickly conduct subsequent drug molecule research and analysis based on the screened target drug molecule.

[0086] The electronic device for drug molecule processing provided in this application can be various types of terminal devices or servers. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0087] Taking servers as an example, such as server clusters deployed in the cloud, AI as a Service (AIaaS) is offered to users. The AIaaS platform breaks down several common AI services and provides them as independent or packaged services in the cloud. This service model is similar to an AI-themed marketplace, where all users can access and use one or more AI services provided by the AIaaS platform through application programming interfaces.

[0088] For example, one type of AI cloud service could be a drug molecule processing service, where a cloud server encapsulates the drug molecule processing program provided in this application embodiment. Users invoke the drug molecule processing service in the cloud service via a terminal (running a client, such as a drug screening client), causing the cloud-deployed server to call the encapsulated drug molecule processing program. Based on the activity information and molecular docking information of candidate drug molecules, a target drug molecule is screened from multiple candidate drug molecules. Subsequently, the target drug molecule responds to a screening request. For example, in a drug screening application, a compound library targeting GPCRs is obtained. Based on the activity information and molecular docking information of candidate drug molecules in the compound library, a target drug molecule targeting GPCRs is screened from multiple candidate drug molecules. Since G protein-coupled receptors have a certain effect on reproductive system syndromes, drug developers can quickly obtain drug molecules with positive effects on reproductive system syndromes from the target drug molecule.

[0089] See Figure 1 , Figure 1 This is a schematic diagram of the application scenario of the medical system 10 provided in the embodiments of this application. The terminal 200 is connected to the server 100 through the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0090] Terminal 200 (running a client, such as a drug screening client) can be used to obtain target proteins and compound libraries (including multiple candidate drug molecules). For example, drug developers can input the target protein (e.g., GPCR) and the compound library (e.g., ZINC library) for drug screening through the input interface of Terminal 200 to obtain the target protein and compound library, and automatically generate screening requests for drug molecules.

[0091] In some embodiments, a drug molecule processing plugin may be embedded in the client running on the terminal to implement an AI-based drug molecule processing method locally on the client. For example, after the terminal 200 receives a screening request for drug molecules (including a compound library for the target protein), it calls the drug molecule processing plugin to implement an AI-based drug molecule processing method. Based on the activity information and molecular docking information of candidate drug molecules, the target drug molecule is screened from multiple candidate drug molecules, and then the screening request for drug molecules is responded to based on the target drug molecule.

[0092] In some embodiments, after receiving a screening request for a drug molecule, the terminal 200 calls the drug molecule processing interface of the server 100 (which can be provided as a cloud service, i.e., a drug molecule processing service). Based on the activity information and molecular docking information of the candidate drug molecules, the server 100 screens out the target drug molecule from multiple candidate drug molecules. Subsequently, it responds to the screening request for the drug molecule based on the target drug molecule. For example, in a drug screening application, drug developers input the target protein (e.g., GPCR) and a compound library (e.g., a ZINC library) for drug screening through the input interface of the drug screening client to obtain the target protein and the compound library. The terminal automatically generates a screening request for the drug molecule and calls the drug molecule processing interface of the server 100. Based on the activity information and molecular docking information of the candidate drug molecules in the compound library, the terminal screens out the target drug molecule targeting the GPCR from multiple candidate drug molecules. Since G protein-coupled receptors have a certain effect on reproductive system syndromes, drug developers can quickly obtain drug molecules with positive effects on reproductive system syndromes from the target drug molecules.

[0093] The structure of the electronic device for drug molecule processing provided in the embodiments of this application is described below. See also... Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device 500 for drug molecule processing provided in an embodiment of this application. The electronic device 500 is described using a server as an example. Figure 2The illustrated electronic device 500 for processing organic samples includes at least one processor 510, a memory 550, at least one network interface 520, and a user interface 530. The various components in the electronic device 500 are coupled together via a bus system 540. It is understood that the bus system 540 is used to enable communication between these components. In addition to a data bus, the bus system 540 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 540.

[0094] The processor 510 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0095] Memory 550 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 550 described in this application embodiment is intended to include any suitable type of memory. Memory 550 may optionally include one or more storage devices physically located away from processor 510.

[0096] In some embodiments, memory 550 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0097] Operating system 551 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0098] The network communication module 552 is used to reach other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0099] In some embodiments, the drug molecule processing device provided in this application can be implemented in software, for example, it can be a drug molecule processing plugin in the terminal mentioned above, or a drug molecule processing service in the server mentioned above. Of course, it is not limited to this; the drug molecule processing device provided in this application can be provided in various software embodiments, including various forms such as applications, software, software modules, scripts, or code.

[0100] Figure 2 A drug molecule processing device 555 stored in a memory 550 is shown. It may be software in the form of programs and plug-ins, such as a drug molecule processing plug-in, and includes a series of modules, including a determination module 5551, a prediction module 5552, a processing module 5553, and a screening module 5554. The determination module 5551, the prediction module 5552, the processing module 5553, and the screening module 5554 are used to implement the drug molecule processing function provided in the embodiments of this application.

[0101] As mentioned above, the artificial intelligence-based organic sample processing method provided in this application can be implemented by various types of electronic devices. See also Figure 3 , Figure 3 This is a flowchart illustrating the artificial intelligence-based organic sample processing method provided in this application embodiment, combined with... Figure 3 The steps shown are explained.

[0102] In the following steps, the target protein refers to a protein that has a role or effect on a certain disease, such as the G protein-coupled receptor GPR54 / Kiss-1, which has a certain role in reproductive system syndrome.

[0103] In step 101, multiple candidate drug molecules targeting the target protein are identified.

[0104] As an example of obtaining candidate drug molecules, the user inputs the target protein (e.g., GPCR) and a compound library (e.g., ZINC library) for drug screening through the terminal's input interface to obtain the target protein and the compound library. The system automatically generates a screening request for drug molecules and sends it to the server. The server parses the screening request for drug molecules, obtains the target protein and the compound library, and reads multiple candidate drug molecules for the target protein from the compound library. This allows for subsequent screening based on multiple candidate drug molecules to obtain the target drug molecule that is effective against the target protein.

[0105] In some embodiments, identifying multiple candidate drug molecules targeting a target protein includes: screening compounds in a compound library based on the target protein to obtain multiple screened compounds; pre-treating the multiple screened compounds and using the pre-treated compounds as candidate drug molecules targeting the target protein.

[0106] For example, since the compound library contains a huge number of compounds, it requires a lot of computation. Therefore, the compounds in the compound library can be coarsely screened based on the target protein to obtain candidate drug molecules that may have an effect on the target protein, thereby reducing the number of candidate drug molecules and improving the efficiency of subsequent drug screening.

[0107] In some embodiments, the compounds in the compound library are screened based on the target protein to obtain multiple screened compounds, including: screening the compounds in the compound library based on molecular druggation rules to obtain multiple compounds that conform to the druggation rules; removing duplicates from the multiple compounds that conform to the druggation rules, and using the deduplicated compounds as the screened compounds.

[0108] For example, compounds in a compound library are first screened based on molecular druggender rules to obtain compounds that meet the druggender rules, i.e., compounds with pharmacological or biological activity. Then, duplicates are removed from these druggender rules to further reduce the number of compounds. The molecular druggender rules are used to assess whether a compound can be used as a drug molecule (not all compounds can be drug molecules). For example, the molecular druggender rules are: 250 ≤ molecular weight ≤ 750; -2 ≤ logarithm of the oil-water partition coefficient ≤ 7; hydrogen bond acceptor + hydrogen bond donor < 10; number of rotatable bonds < 10; topological polar surface area < 150.

[0109] In some embodiments, pretreatment of multiple screened compounds and using the pretreated compounds as candidate drug molecules against the target protein includes: chemically filtering the multiple screened compounds based on the target group to obtain multiple filtered compounds; removing enantiomers of chiral compounds from the multiple filtered compounds, and using the enantiomer-removed compounds as candidate drug molecules against the target protein.

[0110] In this context, the target group refers to an unwanted group in medicinal chemistry, such as molecules containing toxicophore groups and reactive groups. This means removing compounds containing the target group from the screened list, thus performing medicinal chemistry filtering. Since the compound library contains many enantiomers of chiral compounds—compounds with the same molecular weight and structure but reversed left-right arrangement, meaning they have similar functions—enantiomers of chiral compounds can be removed to reduce the number of candidate drug molecules and the subsequent computational workload.

[0111] In step 102, activity prediction processing is performed based on multiple candidate drug molecules and target proteins to obtain activity information for each candidate drug molecule.

[0112] For example, after obtaining multiple candidate drug molecules, the activity of each candidate drug molecule and the target protein is predicted to obtain an activity score (activity information) for each candidate drug molecule. The activity of the candidate drug molecules is evaluated by the activity score so that subsequent drug screening can be carried out based on the activity information.

[0113] See Figure 4 , Figure 4 This is an optional flowchart illustrating an artificial intelligence-based drug molecule processing method provided in an embodiment of this application. Figure 4 Show Figure 3 Step 102 can be implemented through steps 1021-1024: In step 1021, the following processing is performed on any candidate drug molecule among multiple candidate drug molecules: the molecular structure of the candidate drug molecule is encoded to obtain the embedding feature of the candidate drug molecule; in step 1022, the protein structure of the target protein is encoded to obtain the embedding feature of the target protein; in step 1023, the embedding feature of the candidate drug molecule and the embedding feature of the target protein are fused to obtain the activity fusion feature; in step 1024, the activity fusion feature is mapped to obtain the activity information of the candidate drug molecule.

[0114] For example, such as Figure 6 As shown, the molecular structure (molecular graph) of the candidate drug molecule is encoded by the encoder of the prediction network to obtain the embedding features of the candidate drug molecule. The protein structure (protein sequence) of the target protein is encoded to obtain the embedding features of the target protein. Then, the embedding features of the candidate drug molecule and the embedding features of the target protein are fused, and the fused activity fusion features are mapped to obtain the activity score of the candidate drug molecule. Thus, the activity of the candidate drug molecule is learned through artificial intelligence technology, so that subsequent screening processing based on the accurate activity of the candidate drug molecule can be carried out.

[0115] Following the example above, a molecular map of the candidate drug molecule is constructed based on its molecular structure. The molecular map of the candidate drug molecule is then image encoded using an image encoder (e.g., DMPNN) to obtain the embedding features of the candidate drug molecule, thereby obtaining accurate embedding features of the candidate drug molecule through image encoding.

[0116] Following the example above, based on the protein structure of the target protein, the protein sequence of the target protein is determined; the protein sequence of the target protein is encoded in text form using a text converter (e.g., Doc-to-Vector) to obtain the embedding features of the target protein, thereby obtaining accurate embedding features of the target protein through text encoding.

[0117] Following the above example, the embedding features of the candidate drug molecule and the target protein are fused to obtain the activity fusion features, including the following three methods:

[0118] Method 1: The embedding features of the candidate drug molecule and the embedding features of the target protein are summed, and the result of the summing is used as the active fusion feature.

[0119] Method 2: The embedding features of the candidate drug molecule and the embedding features of the target protein are spliced ​​together, and the splicing result is used as the active fusion feature.

[0120] Method 3: Map the embedding features of the candidate drug molecule and the target protein to obtain an intermediate feature vector containing both the candidate drug molecule and the target protein. Perform an affine transformation on the intermediate feature vector to obtain the activity fusion feature.

[0121] Among them, the fusion methods of Method 1 and Method 2 are relatively simple and can save the computational cost of fusion. The fusion method of Method 3 is more accurate and can perform accurate fusion to obtain accurate activity fusion features, so as to make accurate activity prediction in the future.

[0122] Following the example above, the activity fusion features are mapped to obtain the activity information of the candidate drug molecule. This includes: mapping the activity fusion features to the latent vector space to obtain the latent vector of the activity fusion features; performing nonlinear mapping on the latent vector of the activity fusion features; and using the obtained activity score of the candidate drug molecule as the activity information of the candidate drug molecule.

[0123] For example, by mapping the activity fusion features to the latent vector space through the fully connected layer in the prediction network, the latent vector of the activity fusion features is obtained. Then, the latent vector of the activity fusion features is processed by nonlinear mapping through the activation layer in the prediction network, and the activity score (activity score) of the candidate drug molecule is used as the activity information of the candidate drug molecule.

[0124] In step 103, the target protein is subjected to homology modeling to obtain a reference protein with a homologous structure to the target protein.

[0125] Since there is no information on the crystal structure of the target protein (e.g., GPR54 protein), molecular docking cannot be performed based on the target protein. Therefore, it is necessary to use homology modeling to establish a reference protein with a homologous structure to the target protein, that is, the sequence information of a protein with a homologous structure to the target protein.

[0126] In some embodiments, homology modeling is performed on the target protein to obtain a reference protein with a homologous structure to the target protein, including: performing the following processing on any candidate protein in the protein library: performing similarity processing on the sequence of the candidate protein and the sequence of the target protein to obtain the similarity between the candidate protein and the target protein; when the similarity is greater than the similarity threshold, performing structural optimization processing based on the three-dimensional structure of the candidate protein, and using the result of the structural optimization processing as a reference protein with a homologous structure to the target protein.

[0127] For example, if two proteins have similar sequences, then their structures are also similar; that is, two protein sequences with homology (homologous structures) have similar structures. First, any protein (a protein with a known structure) from a protein library (including multiple candidate proteins) is used as a candidate protein (template protein). The sequence of the candidate protein is compared with the sequence of the target protein to obtain the similarity between the candidate and target proteins. When the similarity is greater than a similarity threshold, structural optimization is performed based on the three-dimensional structure of the candidate protein. Specifically, a main-chain structure model of the target protein is constructed using the candidate protein as a prototype. Vacancies are identified in the comparison between the candidate and target proteins, and loop regions are used to model the complete main-chain structure. The side chains of the model are then constructed and optimized to achieve optimization of the entire structural model, thereby obtaining a reference protein sequence with a homologous structure to the target protein.

[0128] In step 104, molecular docking is performed based on the reference protein and multiple candidate drug molecules to obtain molecular docking information for each candidate drug molecule.

[0129] After obtaining the structure of the reference protein, molecular docking is performed on the reference protein and each candidate drug molecule to obtain the molecular docking score of each candidate drug molecule with the reference protein. The molecular docking score of the candidate drug molecule with the reference protein is used as the molecular docking information of the candidate drug molecule. The binding inhibition effect of the candidate drug molecule is evaluated based on the molecular docking information so that the drug can be accurately screened in the future.

[0130] See Figure 5 , Figure 5This is an optional flowchart illustrating an artificial intelligence-based drug molecule processing method provided in an embodiment of this application. Figure 5 Show Figure 3 Step 104 can be implemented through steps 1041-1043: In step 1041, molecular dynamics simulation is performed based on the reference protein to obtain the active site and binding pocket of the reference protein; in step 1042, multiple candidate drug molecules are preprocessed to obtain the molecular conformation of each candidate drug molecule; in step 1043, the following processing is performed on the molecular conformation of each candidate drug molecule: molecular docking scoring is performed based on the active site, binding pocket of the reference protein and the molecular conformation of the candidate drug molecule, and the molecular docking scoring result is used as the molecular docking information of the candidate drug molecule.

[0131] For example, by simulating the structure of a ligand-free reference protein using molecular dynamics, the reference protein is inserted into a polymorphic bilayer (POPE) structure, and the simulated space is filled with water molecules. A 0.15 M concentration of NaCl salt ions is added, and the simulation is performed at 300 K under constant pressure for more than 100 ns. The reference protein in the simulated trajectory during the last 80 ns of this simulation is clustered, and the final cluster center structure is taken as the acceptor structure (active site) for molecular docking. Using the FTMap method (molecular docking method), a series of chemical groups or molecules (e.g., benzene, isobutanol, uric acid, and ethanol) are docked to the pocket region. Regions enriched with multiple chemical molecules constitute the pocket region (binding pocket) for molecular docking in the next step.

[0132] Following the example above, multiple candidate drug molecules are preprocessed separately, including: performing format conversion on each candidate drug molecule to obtain a converted format for each candidate drug molecule; constructing a three-dimensional conformation for each candidate drug molecule based on its converted format; determining the positions where hydrogen atoms can be added based on the three-dimensional conformation of each candidate drug molecule; and adding hydrogen atoms to the positions where hydrogen atoms can be added to obtain the molecular conformation of the candidate drug molecule.

[0133] For example, converting candidate drug molecules from SMILES format to PDB / MOL2 / PDBQT / SDF formats; constructing a three-dimensional conformation of the candidate drug molecule based on the converted format; determining the positions of hydrogen atoms that can be added to the candidate drug molecule (the basic positions), and adding hydrogen atoms.

[0134] In step 105, based on the activity information and molecular docking information of each candidate drug molecule, multiple candidate drug molecules are screened, and the screened candidate drug molecules are used as target drug molecules.

[0135] For example, after obtaining the activity information and molecular docking information of each candidate drug molecule, drug screening is carried out based on the activity information and molecular docking information of each candidate drug molecule to obtain the target drug molecule, thereby ensuring the activity and binding inhibition of the target drug molecule, providing reference guidance for subsequent drug molecule research and shortening the drug research cycle.

[0136] In some embodiments, multiple candidate drug molecules are screened based on the activity information and molecular docking information of each candidate drug molecule, and the screened candidate drug molecules are used as target drug molecules. This includes: clustering multiple candidate drug molecules to obtain multiple drug category sets; and screening candidate drug molecules that meet the requirements of activity information and molecular docking information from the multiple drug category sets to use as target drug molecules.

[0137] For example, candidate drug molecules can be clustered to obtain multiple drug category sets (clusters). Screening can then be performed based on these drug category sets to identify candidate drug molecules that meet both activity and molecular docking information requirements, which can then be used as target drug molecules. Clustering processes result in drug molecules with diverse structural frameworks, thereby increasing the diversity of target drug molecules.

[0138] Following the example above, in each drug category set, the candidate drug molecule with the highest activity information or the highest molecular docking information can be selected as the target drug molecule.

[0139] In some embodiments, selecting candidate drug molecules that meet the requirements for activity information and molecular docking information from multiple drug category sets as target drug molecules includes: performing the following processing on any drug category set of multiple drug category sets: selecting the candidate drug molecule with the highest activity information in the drug category set as the drug molecule to be screened; performing a weighted summation of the activity information, molecular docking information, and drug properties of the drug molecule to be screened to obtain comprehensive drug information of the drug molecule to be screened; and based on the comprehensive drug information of multiple drug molecules to be screened, sorting the multiple drug molecules to be screened in descending order, and selecting the top-ranked drug molecules to be screened as target drug molecules.

[0140] For example, in each drug category set, candidate drug molecules with the highest activity information are selected as the drug molecules to be screened. The activity information, molecular docking information, and drug properties (including solubility, blood-brain barrier permeability, toxicity, etc.) of the drug molecules to be screened are determined, and the weights of the activity information, molecular docking information, and drug properties are determined. Based on the weights of the activity information, molecular docking information, and drug properties, the activity information, molecular docking information, and drug properties of the drug molecules to be screened are weighted and summed to obtain the comprehensive drug information of the drug molecules to be screened. Based on the comprehensive drug information of multiple drug molecules to be screened, the multiple drug molecules to be screened are sorted in descending order, and the top-ranked drug molecules are selected as target drug molecules. In this way, target drug molecules with strong activity, strong binding inhibition ability, and safety are screened from a large number of candidate drug molecules, providing guidance for subsequent drug molecule research and shortening the drug research cycle.

[0141] The following will describe an exemplary application of the embodiments of this application in a real medical application scenario.

[0142] Polycystic ovary syndrome (PCOS) is a hormonal imbalance, the exact cause of which is currently unknown. The ovaries produce estrogen and progesterone, which are female hormones. They also produce androgens, which are male hormones. In PCOS, the ovaries produce too many androgens, causing a hormonal imbalance. This imbalance can have a wide range of effects on the body, ranging from mild to severe, affecting approximately one in ten women. Like any chronic condition, PCOS can impact quality of life to some extent, but if left untreated, it carries the risk of other health problems, including heart disease, diabetes, obesity, endometrial cancer, and infertility.

[0143] In related technologies, the treatment of PCOS mainly involves hormone therapy to increase estrogen levels or reduce the effects of elevated androgens. Currently, there are no drugs on the market specifically targeting polycystic ovary syndrome (PCOS). Figure 7 As shown, abnormal MAPK / ERK signaling pathways lead to metabolic signal defects and excessive ovarian androgen secretion in PCOS patients. G protein-coupled receptor GPR54 / Kiss-1 plays a role in reproductive system syndromes. The expression levels of several key proteins in the ovary can be significantly affected by the binding of the polypeptide molecule Kisspeptin (a type of hormone produced by Kp neurons that regulates reproductive activity by regulating estrogen levels in the body) and GPR54 protein, thereby alleviating polycystic ovary syndrome.

[0144] In implementing the embodiments of this application, the applicant discovered that although Kisspeptin can bind to GPR54 to alleviate PCOS, its high expression in the central nervous system prevents direct injection of Kisspeptin from achieving therapeutic effects. Therefore, finding a small molecule that can selectively bind to the GPR54 protein, while also avoiding cross-brain barriers and toxic side effects on the central nervous system, is of great research potential and clinical value.

[0145] However, very few drug molecules have been developed targeting GPR54, mainly for the following reasons: 1) The crystal structure of target GPR54 is unknown, making structure-based drug molecule design difficult; 2) Data on related active molecules is relatively scarce. For example... Figure 8 As shown, by designing, synthesizing, testing, and redesigning the molecule, the structure-activity relationship was optimized to find a lead compound with ideal activity (lead compound 1) with an activity of 1.2 μM.

[0146] However, the aforementioned drug design strategies require significant human and material resources for continuous trial and error, followed by optimization. This process is not only inefficient and time-consuming but also has relatively low accuracy. Furthermore, it necessitates ongoing testing of blood-brain barrier crossing rates and other ADMET properties to ensure the safety, physicochemical properties, and in vivo metabolic stability of the drug molecules.

[0147] To address the aforementioned issues, this application provides an artificial intelligence-based virtual drug screening method for polycystic ovary syndrome (PCOS). This method uses artificial intelligence to screen 50 potential small molecule drugs (target drug molecules) that inhibit GPR54 (the target protein). In other words, it uses artificial intelligence to screen drug molecules that inhibit PCOS, providing guidance for the discovery of lead compounds for the treatment of PCOS. This eliminates the need for extensive trial-and-error processes and improves the efficiency of molecule screening.

[0148] This application first analyzes the amino acid sequence information of the GPR54 target and establishes a docking model for further evaluation of small molecule activity using homology modeling. Then, an artificial intelligence algorithm is used to learn the activity of small molecules targeting all GPCRs, effectively predicting and scoring the activity of the ZINC molecule library. Finally, the ADMET model is used to predict the properties of each small molecule, thus ensuring molecular activity as much as possible while allowing for screening based on the ADMET properties of the molecules.

[0149] The AI-based virtual drug screening method for polycystic ovary syndrome (PCOS) comprises eight parts: selection of the compound library, preprocessing of the compound library, deep learning-based compound activity scoring, molecular docking model establishment, molecular docking scoring, post-processing screening, ADMET property prediction of compounds, and final compound screening. The processing flow of the AI-based virtual drug screening method for PCOS is detailed below:

[0150] 1) Selection of the compound library to be screened

[0151] Considering the time and precision of virtual molecular screening, a sub-library of small molecules with in-vitro activity from the ZINC library was selected, totaling 276,003 small molecules.

[0152] 2) Preprocessing of the compound library

[0153] First, the selected compound library is preprocessed, mainly including the following processes:

[0154] Step 1: Screening based on Lipinski's Rule of Five (an empirical rule for assessing whether a compound can be used as a drug, or whether a compound with pharmacological or biological activity can be used as an oral drug). Specific screening criteria are as follows:

[0155] • 250 ≤ Molecular Weight (MW) ≤ 750

[0156] • -2 ≤ Logarithm of the oil-water partition coefficient (clogP) of the compound ≤ 7

[0157] • Hydrogen bond acceptor (HBA) + hydrogen bond donor (HBD) < 10

[0158] • Number of rotatable bonds < 10

[0159] • Topological polar surface area (TPSA) < 150

[0160] Step 2: Remove duplicates;

[0161] Step 3, Medicinal Chemistry Filter (MCF): Used to remove molecules containing unwanted groups (target groups, such as those containing toxicophores and reactive groups) in medicinal chemistry;

[0162] Step 4: Remove enantiomers of chiral compounds (retain only one SMILES structure): Since the compound library contains many enantiomers of chiral compounds (compounds with the same molecular weight and molecular structure but with opposite left and right arrangements) (among molecules that constitute a chiral relationship, one is called the enantiomer of the other), this step can remove most of the small molecules.

[0163] In step 1, 260,300 molecules were obtained; in step 2, 160,455 small molecules without repetition were obtained; and in steps 3 and 4, 18,907 molecules were obtained.

[0164] 3) Deep learning-based compound activity scoring (DL activity scoring)

[0165] like Figure 9 As shown, using the sequences of all known GPCR proteins and the SMILES of known small molecules (candidate drug molecules) with binding activity to GPCRs as input, the GPCR protein sequences are used to obtain protein embedding features based on a text converter (e.g., Doc-to-Vector), and the small molecules with binding activity to GPCRs are used to obtain small molecule embedding features based on a graph encoder (e.g., DMPNN). Then, a fully connected layer is used to predict the binding strength between small molecules and GPCR proteins based on the small molecule embedding features and the protein embedding features. Figure 9 The model shown performed a deep learning activity prediction on the 18,907 molecules selected above, and obtained the activity scores (represented by the negative logarithm of the IC50 value, pIC50) for each of the 18,907 molecules.

[0166] 4) Establishment of molecular docking model

[0167] A) Homogeneous modeling of GPR54

[0168] Since there is no information on the crystal structure of the GPR54 protein, a model needs to be built using homology modeling. The structures obtained through homology modeling include those similar to 7TM.

[0169] B) Active amino acids (active sites) and pocket detection

[0170] Most GPCR proteins contain multiple binding pockets (containing active sites), and here we choose the binding site located at 7TMhelices.

[0171] Since the exact location of the ligand's active site or binding pocket is unknown, molecular dynamics (MD) simulations (a comprehensive molecular simulation method combining physics, mathematics, and chemistry) are needed to predict and confirm the small molecule binding site and which active amino acids might constitute the binding pocket. By simulating the ligand-free protein structure using molecular dynamics, the protein is inserted into a polymorphic bilayer (POPE) structure, and the simulated space is filled with water molecules. A 0.15M concentration of NaCl salt ions is added, and the simulation is performed at 300K under constant pressure for more than 100 ns. The protein structures of the simulated trajectories during the last 80 ns of the simulation are clustered, and the final cluster center structure is taken as the acceptor structure for molecular docking. Using the FTMap method (molecular docking method), a series of chemical groups or molecules (e.g., benzene, isobutanol, uric acid, and ethanol) are docked to the pocket region. Regions enriched with multiple chemical molecules constitute the pocket region for molecular docking in the next step.

[0172] 5) Molecular docking scoring

[0173] The input file is prepared for the docking software (AutoDock Vina). This process requires determining the correct molecular conformation for each molecule in the compound library. The atomic charges and hydrogen bonds are crucial for successful virtual screening. Hydrogen bonds significantly increase the binding affinity between small molecules and proteins; therefore, the presence of a protonated hydrogen atom bonded to an atom at a specific position on the protein is essential. The following are the processing steps for molecules in the compound library:

[0174] a. Convert small molecules from SMILES format to PDB / MOL2 / PDBQT / SDF formats;

[0175] b. Generation of 3D conformations of small molecules;

[0176] c. Determine the position of the small molecule where a hydrogen atom can be added (the basic position), and add a hydrogen atom;

[0177] d. Calculate the charge number;

[0178] e. Set the correct atom type.

[0179] In this process, the processed small molecules are combined with the structures obtained from homology modeling to determine the value of energy reduction. The greater the reduction value, the higher the molecular docking score and the stronger the binding.

[0180] 6) Post-processing screening

[0181] For molecules scored by the two models, clustering was performed. In order to obtain molecules with structurally diverse skeletons, the above 18,000 small molecules were clustered, resulting in 6,446 clusters. Then, the molecules with the highest DL activity scores were selected from each of the 6,446 clusters.

[0182] Clustering is a similarity-based clustering method that aims to increase molecular diversity and reduce similar molecules. A similarity threshold is set so that molecules with a similarity greater than 0.6 are clustered together.

[0183] 7) Prediction of ADMET properties of compounds (drug absorption, partition, metabolism, excretion, and toxicity)

[0184] This study selected several fundamental properties of drug molecules, including blood-brain barrier permeability (BBBP), kinetic solubility (S, which measures the solubility of a drug molecule in the body), and toxicity. The 18,907 compounds obtained were then tagged with five indicators: DL activity score, docking score, solubility, BBBP permeability, and toxicity. Subsequent compound selection will primarily combine these five indicators for comprehensive selection.

[0185] 8) Final compound screening

[0186] The 6446 molecules obtained from post-processing screening were ranked from high to low according to their DL activity score (pIC50). Then, considering solubility, blood-brain barrier permeability, toxicity, and docking score, the following were selected: Figures 10A-10B The potential active molecules (target drug molecules) shown include 50 compounds. The properties are referenced as follows: solubility log(S, mol / L) is selected as greater than -4.5, BBBP is selected as a probability value less than 0.5, and no obvious predicted toxicity was observed.

[0187] In summary, the artificial intelligence-based virtual drug screening method for polycystic ovary syndrome (PCOS) proposed in this application uses artificial intelligence algorithms to screen 50 potential drug molecules with inhibitory activity for PCOS, a chronic disease with a high incidence rate. This expands the diversity of chemical structures and provides lead compounds for further development targeting this disease.

[0188] This concludes the description of the artificial intelligence-based drug molecule processing method provided in this application, using the exemplary application and implementation of the server provided in the embodiments of this application. This application also provides a drug molecule processing device. In practical applications, the functional modules in the drug molecule processing device can be collaboratively implemented using the hardware resources of electronic devices (such as terminal devices, servers, or server clusters), such as computing resources like processors, communication resources (such as those supporting various communication methods like optical fiber and cellular), and memory. Figure 2 A drug molecule processing device 555 stored in memory 550 is shown. It can be software in the form of programs and plug-ins, such as software modules designed in programming languages ​​such as C / C++ and Java, application software designed in programming languages ​​such as C / C++ and Java, or dedicated software modules, application programming interfaces, plug-ins, cloud services, etc. in large software systems. Examples of different implementation methods are given below.

[0189] Example 1: Drug molecule processing device is a mobile application and module.

[0190] The drug molecule processing device 555 in this embodiment can provide a software module designed using programming languages ​​such as C / C++ and Java, which can be embedded into various mobile applications based on systems such as Android or iOS (stored as executable instructions in the storage medium of the mobile device and executed by the processor of the mobile device). This allows the device to directly use its own computing resources to complete the relevant drug molecule screening tasks, and to periodically or irregularly transmit the processing results to a remote server through various network communication methods, or to save them locally on the mobile device.

[0191] Example 2: Drug molecule processing device is a server application and platform.

[0192] The drug molecule processing device 555 in this embodiment can be provided as a dedicated software module in an application software or large software system designed using programming languages ​​such as C / C++ and Java. It runs on the server side (stored in the server-side storage medium as executable instructions and run by the server-side processor). The server uses its own computing resources to complete the relevant drug molecule screening tasks.

[0193] This application embodiment can also provide a drug molecule platform (for drug molecule screening) for use by individuals, groups or organizations by mounting a customized, easy-to-interact web interface or other user interfaces on a distributed, parallel computing platform composed of multiple servers.

[0194] Example 3: The drug molecule processing device is a server-side application programming interface (API) and plugin.

[0195] The drug molecule processing device 555 in this embodiment can be provided as a server-side API or plugin for users to call, to execute the AI-based drug molecule processing method of this embodiment, and to be embedded in various applications.

[0196] Example 4: Drug molecule processing devices are mobile device client APIs and plugins.

[0197] The drug molecule processing device 555 in this application embodiment can be provided as an API or plugin on a mobile device for users to call in order to execute the artificial intelligence-based drug molecule processing method of this application embodiment.

[0198] Example 5: The drug molecule processing device is an open service in the cloud.

[0199] The drug molecule processing device 555 in this application embodiment can provide a drug molecule processing cloud service developed for users, allowing individuals, groups or organizations to obtain target drug molecules.

[0200] The drug molecule processing device 555 includes a series of modules, including a determination module 5551, a prediction module 5552, a processing module 5553, and a screening module 5554. The following description further illustrates how the various modules in the drug molecule processing device 555 provided in this embodiment cooperate to implement the drug molecule processing scheme.

[0201] The module 5551 is used to identify multiple candidate drug molecules targeting the target protein; the prediction module 5552 is used to perform activity prediction processing based on the multiple candidate drug molecules and the target protein to obtain activity information for each candidate drug molecule; the processing module 5553 is used to perform homology modeling processing on the target protein to obtain a reference protein with a homologous structure to the target protein; and to perform molecular docking processing based on the reference protein and the multiple candidate drug molecules to obtain molecular docking information for each candidate drug molecule; the screening module 5554 is used to screen the multiple candidate drug molecules based on the activity information and molecular docking information of each candidate drug molecule, and to select the screened candidate drug molecules as target drug molecules.

[0202] In some embodiments, the determining module 5551 is further configured to screen compounds in the compound library based on the target protein to obtain a plurality of screened compounds; preprocess the plurality of screened compounds and use the preprocessed compounds as candidate drug molecules against the target protein.

[0203] In some embodiments, the determining module 5551 is further configured to perform molecular drug-making rule-based screening on compounds in the compound library based on the target protein to obtain multiple compounds that conform to the drug-making rules; and to perform deduplication on the multiple compounds that conform to the drug-making rules, and use the deduplicated compounds as the screened compounds.

[0204] In some embodiments, the determining module 5551 is further configured to perform chemical filtration on the plurality of screened compounds based on the target group to obtain a plurality of filtered compounds; remove enantiomers of chiral compounds from the plurality of filtered compounds, and use the compounds from which the enantiomers are removed as candidate drug molecules for the target protein.

[0205] In some embodiments, the prediction module 5552 is further configured to perform the following processing on any one of the plurality of candidate drug molecules: encoding the molecular structure of the candidate drug molecule to obtain the embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain the embedding feature of the target protein; fusing the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; and mapping the activity fusion feature to obtain the activity information of the candidate drug molecule.

[0206] In some embodiments, the prediction module 5552 is further configured to construct a molecular map of the candidate drug molecule based on the molecular structure of the candidate drug molecule; and to perform image encoding processing on the molecular map of the candidate drug molecule to obtain the embedding features of the candidate drug molecule.

[0207] In some embodiments, the prediction module 5552 is further configured to determine the protein sequence of the target protein based on the protein structure of the target protein; and to perform text conversion processing on the protein sequence of the target protein to obtain the embedding features of the target protein.

[0208] In some embodiments, the prediction module 5552 is further configured to sum the embedding features of the candidate drug molecule and the embedding features of the target protein, and use the result of the summing process as the activity fusion feature; or, to splice the embedding features of the candidate drug molecule and the embedding features of the target protein, and use the result of the splicing process as the activity fusion feature.

[0209] In some embodiments, the prediction module 5552 is further configured to perform mapping processing on the embedding features of the candidate drug molecule and the embedding features of the target protein to obtain an intermediate feature vector containing the candidate drug molecule and the target protein; and to perform an affine transformation on the intermediate feature vector to obtain the activity fusion feature.

[0210] In some embodiments, the prediction module 5552 is further configured to map the activity fusion feature to a latent vector space to obtain the latent vector of the activity fusion feature; perform nonlinear mapping processing on the latent vector of the activity fusion feature, and use the obtained activity score of the candidate drug molecule as the activity information of the candidate drug molecule.

[0211] In some embodiments, the processing module 5553 is further configured to perform the following processing on any candidate protein in the protein library: perform similarity processing on the sequence of the candidate protein and the sequence of the target protein to obtain the similarity between the candidate protein and the target protein; when the similarity is greater than a similarity threshold, perform structure optimization processing based on the three-dimensional structure of the candidate protein, and use the result of the structure optimization processing as a reference protein having a homologous structure with the target protein.

[0212] In some embodiments, the processing module 5553 is further configured to perform molecular dynamics simulation based on the reference protein to obtain the active site and binding pocket of the reference protein; preprocess the plurality of candidate drug molecules to obtain the molecular conformation of each candidate drug molecule; and perform the following processing on the molecular conformation of each candidate drug molecule: perform molecular docking scoring based on the active site of the reference protein, the binding pocket and the molecular conformation of the candidate drug molecule, and use the result of the molecular docking scoring as the molecular docking information of the candidate drug molecule.

[0213] In some embodiments, the processing module 5553 is further configured to perform format conversion processing on the plurality of candidate drug molecules respectively to obtain a conversion format for each candidate drug molecule; construct a three-dimensional conformation of each candidate drug molecule based on the conversion format of each candidate drug molecule; determine the positions where hydrogen atoms can be added based on the three-dimensional conformation of each candidate drug molecule; and add hydrogen atoms to the positions where hydrogen atoms can be added to obtain the molecular conformation of the candidate drug molecule.

[0214] In some embodiments, the screening module 5554 is further configured to perform clustering processing on the plurality of candidate drug molecules to obtain a plurality of drug category sets; and to screen out candidate drug molecules that meet the requirements of activity information and molecular docking information from the plurality of drug category sets as the target drug molecules.

[0215] In some embodiments, the screening module 5554 is further configured to perform the following processing on any one of the plurality of drug category sets: selecting the candidate drug molecule with the highest activity information in the drug category set as the drug molecule to be screened; performing a weighted summation of the activity information, molecular docking information, and drug properties of the drug molecule to be screened to obtain the comprehensive drug information of the drug molecule to be screened; and based on the comprehensive drug information of the plurality of drug molecules to be screened, performing a descending sorting process on the plurality of drug molecules to be screened, and selecting the portion of drug molecules to be screened that are ranked first as the target drug molecule.

[0216] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the artificial intelligence-based drug molecule processing method described above in this application.

[0217] This application provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to execute the AI-based question-answering method provided in this application. For example... Figure 3-5 The method for processing drug molecules based on artificial intelligence is shown.

[0218] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0219] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0220] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0221] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0222] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. An artificial intelligence-based drug molecule processing method, characterized by, The method comprises: determining a plurality of candidate drug molecules for a target protein; performing the following processing on any candidate drug molecule in the plurality of candidate drug molecules: encoding the molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain an embedding feature of the target protein; performing fusion processing on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; performing mapping processing on the activity fusion feature to obtain activity information of the candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, performing screening processing on the plurality of candidate drug molecules, and selecting a candidate drug molecule obtained by screening as a target drug molecule.

2. The method of claim 1, wherein, The method comprises: determining a plurality of candidate drug molecules for a target protein; performing the following processing on any candidate drug molecule in the plurality of candidate drug molecules:

3. The method of claim 2, wherein, encoding the molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain an embedding feature of the target protein; performing fusion processing on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; performing mapping processing on the activity fusion feature to obtain activity information of the candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; 4. The method of claim 2, wherein, performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, performing screening processing on the plurality of candidate drug molecules, and selecting a candidate drug molecule obtained by screening as a target drug molecule. The method comprises:

5. The method of claim 1, wherein, determining a plurality of candidate drug molecules for a target protein; performing the following processing on any candidate drug molecule in the plurality of candidate drug molecules: encoding the molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain an embedding feature of the target protein; 6. The method of claim 1, wherein, performing fusion processing on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; performing mapping processing on the activity fusion feature to obtain activity information of the candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; 7. The method of claim 1, wherein, based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, performing screening processing on the plurality of candidate drug molecules, and selecting a candidate drug molecule obtained by screening as a target drug molecule. The method comprises: determining a plurality of candidate drug molecules for a target protein; performing the following processing on any candidate drug molecule in the plurality of candidate drug molecules: encoding the molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain an embedding feature of the target protein; performing fusion processing on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; performing mapping processing on the activity fusion feature to obtain activity information of the candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, performing screening processing on the plurality of candidate drug molecules, and selecting a candidate drug molecule obtained by screening as a target drug molecule. The method comprises: determining a plurality of candidate drug molecules for a target protein; performing the following processing on any candidate drug molecule in the plurality of candidate drug molecules: encoding the molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain an embedding feature of the target protein; performing fusion processing on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; performing mapping processing on the activity fusion feature to obtain activity information of the candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, performing screening processing on the plurality of candidate drug molecules, and selecting a candidate drug molecule obtained by screening as a target drug molecule. The method comprises: determining a plurality of candidate drug molecules for a target protein; performing the following processing on any candidate drug molecule in the plurality of candidate drug molecules: encoding the molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule; encoding the protein structure of the target protein to obtain an embedding feature of the target protein; performing fusion processing on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature; performing mapping processing on the activity fusion feature to obtain activity information of the candidate drug molecule; performing homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein; performing molecular docking processing based on the reference protein and the plurality of candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, performing screening processing on the plurality of candidate drug molecules, and selecting a candidate drug molecule obtained by screening as a target drug molecule. The embedding feature of the candidate drug molecule and the embedding feature of the target protein are added, and a result of the adding is taken as the activity fusion feature; or, The embedding feature of the candidate drug molecule and the embedding feature of the target protein are spliced, and a result of the splicing is taken as the activity fusion feature.

8. The method of claim 1, wherein, The embedding feature of the candidate drug molecule and the embedding feature of the target protein are fused to obtain an activity fusion feature, including: The embedding feature of the candidate drug molecule and the embedding feature of the target protein are mapped to obtain an intermediate feature vector containing the candidate drug molecule and the target protein; The intermediate feature vector is subjected to affine transformation to obtain the activity fusion feature.

9. The method of claim 1, wherein, The activity fusion feature is mapped to obtain activity information of the candidate drug molecule, including: The activity fusion feature is mapped to a latent vector space to obtain a latent vector of the activity fusion feature; The latent vector of the activity fusion feature is subjected to nonlinear mapping processing, and an activity score of the candidate drug molecule obtained by the processing is taken as the activity information of the candidate drug molecule.

10. The method of claim 1, wherein, The target protein is subjected to homology modeling processing to obtain a reference protein having a homologous structure with the target protein, including: The following processing is performed on any candidate protein in a protein library: The sequence of the candidate protein is subjected to similarity processing with the sequence of the target protein to obtain a similarity between the candidate protein and the target protein; When the similarity is greater than a similarity threshold, a structure optimization processing is performed based on the three-dimensional structure of the candidate protein, and a result of the structure optimization processing is taken as the reference protein having the homologous structure with the target protein.

11. The method of claim 1, wherein, The reference protein and the plurality of candidate drug molecules are subjected to molecular docking processing to obtain molecular docking information of each of the candidate drug molecules, including: The reference protein is subjected to molecular dynamics simulation processing to obtain an active site and a binding pocket of the reference protein; The plurality of candidate drug molecules are respectively preprocessed to obtain a molecular conformation of each of the candidate drug molecules; The following processing is performed on the molecular conformation of each of the candidate drug molecules: molecular docking scoring is performed based on the active site, the binding pocket of the reference protein, and the molecular conformation of the candidate drug molecule, and a result of the molecular docking scoring is taken as the molecular docking information of the candidate drug molecule.

12. The method of claim 11, wherein, The plurality of candidate drug molecules are respectively preprocessed to obtain a molecular conformation of each of the candidate drug molecules, including: The plurality of candidate drug molecules are respectively subjected to format conversion processing to obtain a converted format of each of the candidate drug molecules; Based on the converted format of each of the candidate drug molecules, a three-dimensional conformation of each of the candidate drug molecules is constructed; Based on the three-dimensional conformation of each of the candidate drug molecules, a position of a hydrogen atom capable of being added is determined; A hydrogen atom is added at the position of the hydrogen atom capable of being added to obtain a molecular conformation of the candidate drug molecule.

13. The method of claim 1, wherein, The screening processing is performed on the multiple candidate drug molecules based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, and a candidate drug molecule screened out is taken as a target drug molecule, and the method comprises the steps of: performing clustering processing on the multiple candidate drug molecules to obtain multiple drug category sets; in the multiple drug category sets, a candidate drug molecule meeting the requirements of activity information and molecular docking information is screened out as the target drug molecule.

14. The method of claim 13, wherein, The screening processing is performed on the multiple candidate drug molecules based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, and a candidate drug molecule screened out is taken as a target drug molecule, and the method comprises the steps of: for any drug category set of the multiple drug category sets, the following processing is performed: a candidate drug molecule with the highest activity information in the drug category set is taken as a drug molecule to be screened; weighted sum processing is performed on the activity information, the molecular docking information and the drug property of the drug molecule to be screened to obtain comprehensive drug information of the drug molecule to be screened; based on the comprehensive drug information of multiple drug molecules to be screened, the multiple drug molecules to be screened are sorted in descending order, and part of the drug molecules to be screened sorted in the front are taken as the target drug molecules.

15. An artificial intelligence-based drug molecule processing apparatus, characterized by, The device comprises: a determination module configured to determine multiple candidate drug molecules for a target protein; a prediction module configured to perform the following processing for any candidate drug molecule of the multiple candidate drug molecules: encoding processing is performed on a molecular structure of the candidate drug molecule to obtain an embedding feature of the candidate drug molecule, and encoding processing is performed on a protein structure of the target protein to obtain an embedding feature of the target protein; fusion processing is performed on the embedding feature of the candidate drug molecule and the embedding feature of the target protein to obtain an activity fusion feature, and mapping processing is performed on the activity fusion feature to obtain activity information of the candidate drug molecule; a processing module configured to perform homology modeling processing on the target protein to obtain a reference protein having a homologous structure with the target protein, and perform molecular docking processing based on the reference protein and the multiple candidate drug molecules to obtain molecular docking information of each of the candidate drug molecules; a screening module configured to perform screening processing on the multiple candidate drug molecules based on the activity information of each of the candidate drug molecules and the molecular docking information of each of the candidate drug molecules, and take a candidate drug molecule screened out as a target drug molecule.

16. The device of claim 15, wherein: the determination module is further configured to perform screening processing on compounds in a compound library based on a target protein to obtain multiple screened compounds; the multiple screened compounds are preprocessed, and the preprocessed compounds are taken as candidate drug molecules for the target protein.

17. The device of claim 16, wherein: the determination module is further configured to perform screening processing on compounds in a compound library based on a target protein based on a molecular drug-making rule to obtain multiple compounds meeting the drug-making rule; The multiple compounds meeting the medicine rule are subjected to a deduplication process, and the compounds after the deduplication process are taken as screened compounds.

18. An electronic device, comprising: The electronic device comprises: a memory for storing executable instructions; a processor for executing the executable instructions stored in the memory to implement the artificial intelligence-based drug molecule processing method in any one of claims 1 to 14.

19. A computer-readable storage medium, characterized in that, The memory stores executable instructions for implementing the artificial intelligence-based drug molecule processing method in any one of claims 1 to 14 when executed by the processor.

20. A computer program product comprising computer-executable instructions or a computer program, characterized in that, The computer executable instructions or computer programs are executed by the processor to implement the artificial intelligence-based drug molecule processing method in any one of claims 1 to 14.

Images