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

Abstract

The present invention provides a drug molecule processing method, device, electronic device and storage medium based on artificial intelligence; the method includes: determining the drug molecule property requirements in response to a generation request for drug molecules; obtaining randomly sampled drug molecule vectors; transforming the randomly sampled drug molecule vectors to obtain drug molecule vectors that meet the drug molecule property requirements; decoding the drug molecule vectors that meet the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector; generating drug molecules corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecule, so as to respond to the generation request. Through the present invention, drug molecules that meet the specific drug molecule property requirements can be obtained to meet the user's demand for drug molecules.

Description

Technical Field

[0001] The present invention relates to intelligent medical technology, and in particular to an artificial intelligence-based drug molecule processing method, device, electronic device and storage medium. Background Art

[0002] Artificial Intelligence (AI) is a comprehensive technology in computer science. By studying the design principles and implementation methods of various intelligent machines, it enables machines to have the functions of perception, reasoning and decision-making. Artificial Intelligence technology is a comprehensive discipline that covers a wide range of fields, such as natural language processing technology and machine learning / deep learning. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0003] By applying artificial intelligence technology to drug research and development, some candidate drug molecules can be obtained through the machine's reasoning and decision-making functions, thereby reducing the time cost of manual drug research and development.

[0004] However, the candidate drug molecules currently obtained often cannot meet the needs of users. For example, when users require candidate drug molecules to meet hydrophilicity, the candidate drug molecules finally obtained do not meet hydrophilicity but meet hydrophobicity. Summary of the invention

[0005] The embodiments of the present invention provide a drug molecule processing method, device, electronic device and storage medium based on artificial intelligence, which can obtain drug molecules that meet the requirements of specific drug molecule properties and meet the needs of users for drug molecules.

[0006] The technical solution of the embodiment of the present invention is achieved as follows:

[0007] An embodiment of the present invention provides a drug molecule processing method based on artificial intelligence, the method comprising:

[0008] In response to a request for generating a drug molecule, determining a drug molecule property requirement;

[0009] Obtain randomly sampled drug molecule vectors;

[0010] Transforming the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements;

[0011] Decoding the drug molecule vector that meets the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector;

[0012] According to the mapping relationship between the grammatical rules and the drug molecules, the drug molecules corresponding to the first grammatical rules are generated to respond to the generation request.

[0013] An embodiment of the present invention provides a drug molecule processing device based on artificial intelligence, the device comprising:

[0014] A first determination module, configured to determine a drug molecule property requirement in response to a generation request for a drug molecule;

[0015] An acquisition module, used to acquire randomly sampled drug molecule vectors;

[0016] A prediction module, used for transforming the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements;

[0017] An encoding module, used for decoding the drug molecule vector that meets the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector;

[0018] The second determination module is used to generate the drug molecule corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecule, so as to respond to the generation request.

[0019] In the above technical solution, the acquisition unit is also used to perform random sampling processing on the set of drug molecule vectors that obey the multivariate Gaussian distribution to obtain randomly sampled drug molecule vectors.

[0020] In the above technical solution, the prediction module is also used to predict the randomly sampled drug molecule vector through a molecular property prediction model to obtain the drug property corresponding to the randomly sampled drug molecule vector;

[0021] The drug property corresponding to the randomly sampled drug molecule vector is compared with the drug molecule property requirement, and the randomly sampled drug molecule vector is updated according to the comparison result to obtain a drug molecule vector that meets the drug molecule property requirement.

[0022] In the above technical solution, the prediction module is also used to predict the randomly sampled drug molecule vector through a molecular property prediction model to obtain a probability value of at least one drug property corresponding to the randomly sampled drug molecule vector;

[0023] Subtracting the probability value of at least one drug attribute corresponding to the randomly sampled drug molecule vector from the probability value of the drug attribute in the drug molecule attribute requirement to obtain a corresponding difference value;

[0024] When the difference is greater than a difference threshold, the randomly sampled drug molecule vector is updated according to the drug molecule property requirement to obtain a drug molecule vector that meets the drug molecule property requirement.

[0025] In the above technical solution, the encoding module is also used to decode the drug molecule vector that meets the drug molecule property requirements through at least one decoder of the encoding model to generate at least one first grammatical rule for the drug molecule vector, and the encoding model includes multiple decoders.

[0026] In the above technical solution, the encoding module is also used to decode the output of the n-1th decoder and the drug molecule vector that meets the drug molecule property requirements through the nth decoder in the encoding model to obtain the first grammatical rule of the drug molecule vector corresponding to the nth decoder;

[0027] Wherein, n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the coding model.

[0028] In the above technical solution, the second determination module is further used to combine at least one first grammar rule to obtain a grammar rule sequence corresponding to the drug molecule vector;

[0029] The drug molecule corresponding to the first grammatical rule is obtained according to the mapping relationship between the grammatical rule sequence and the drug molecule, and the grammatical rule sequence corresponding to the drug molecule vector.

[0030] In the above technical solution, the second determination module is further used to combine at least one first grammar rule to obtain a grammar rule sequence corresponding to the drug molecule vector;

[0031] According to the mapping relationship between the molecular linear input specification and the grammatical rule sequence, and the grammatical rule sequence corresponding to the drug molecular vector, the molecular linear input specification corresponding to the drug molecular vector is obtained;

[0032] The drug molecule corresponding to the first grammar rule is obtained according to the mapping relationship between the molecular linear input specification and the drug molecule, and the molecular linear input specification corresponding to the drug molecule vector.

[0033] In the above technical solution, the device also includes:

[0034] A first training module is used to sample the grammatical rules of the training samples through the mean vector and the covariance matrix in the encoder in the encoding model to obtain the feature vector of the training samples;

[0035] Decoding the feature vector of the training sample by a decoder in the coding model to obtain a second grammar rule corresponding to the training sample;

[0036] The model parameters of the encoding model are updated according to the grammatical rule of the training sample and the second grammatical rule of the training sample.

[0037] In the above technical solution, the device also includes:

[0038] The disassembly module is used to disassemble the grammatical rule sequence corresponding to the molecular linear input specification of the training sample to obtain the grammatical rule of at least one training sample.

[0039] In the above technical solution, the decoder is the nth decoder, and the first training module is further used to decode the output of the n-1th encoder and the feature vector of the training sample through the nth decoder in the coding model to obtain a second grammar rule corresponding to the training sample;

[0040] Wherein, n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the coding model.

[0041] In the above technical solution, the device also includes:

[0042] The second training module is used to predict the drug molecule vector of the training sample through the molecular property prediction model to obtain the drug property corresponding to the training sample;

[0043] Determining the value of the loss function of the molecular property prediction model according to the drug properties of the training sample and the standard drug properties;

[0044] According to the value of the loss function, the model parameters of the molecular property prediction model are updated.

[0045] The embodiment of the present invention provides a drug molecule processing device based on artificial intelligence, comprising:

[0046] A memory for storing executable instructions;

[0047] The processor is used to implement the artificial intelligence-based drug molecule processing method provided by an embodiment of the present invention when executing the executable instructions stored in the memory.

[0048] An embodiment of the present invention provides a storage medium storing executable instructions for causing a processor to execute the instructions to implement the artificial intelligence-based drug molecule processing method provided by the embodiment of the present invention.

[0049] The embodiments of the present invention have the following beneficial effects:

[0050] 1. Perform a series of processing on randomly sampled drug molecules through the molecular property prediction model in the drug molecule model to obtain drug molecule vectors that meet the requirements of drug molecule properties, so as to efficiently obtain drug molecules that meet the requirements of specific drug molecule properties;

[0051] 2. The drug molecule vectors required for the drug molecule properties are decoded and processed by the encoding model in the drug molecule model to obtain the grammatical rules corresponding to the drug molecule vectors, so as to obtain the drug molecules corresponding to the grammatical rules. By introducing the constraints of the grammatical rules, the effectiveness of the drug molecules is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of an application scenario of an artificial intelligence-based drug molecule processing system 10 provided in an embodiment of the present invention;

[0053] Figure 2 is a schematic structural diagram of an artificial intelligence-based drug molecule processing device 500 provided in an embodiment of the present invention;

[0054] Figure 3 is a schematic structural diagram of an artificial intelligence-based drug molecule processing device 555 provided in an embodiment of the present invention;

[0055] Figures 4A-4C is a schematic flow chart of a drug molecule processing method based on artificial intelligence provided by an embodiment of the present invention;

[0056] Figure 5 It is a schematic diagram of the process of drug development provided by an embodiment of the present invention;

[0057] Figure 6 is a schematic diagram of a variational autoencoder provided by an embodiment of the present invention;

[0058] Figure 7 is a schematic diagram of a molecular linear input specification grammar rule set provided by an embodiment of the present invention;

[0059] Figure 8 It is a schematic diagram of conversion between a SMILES expression and a grammar rule sequence provided by an embodiment of the present invention;

[0060] Fig. 9 is a schematic diagram of a molecular property prediction model provided by an embodiment of the present invention;

[0061] Fig.10 is a schematic diagram of obtaining a drug molecular structure provided by an embodiment of the present invention;

[0062] Fig.11 It is a comparative schematic diagram of the experimental effects provided by the embodiments of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0064] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0065] In the following description, the terms "first\second" are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0067] Before further describing the embodiments of the present invention in detail, the nouns and terms involved in the embodiments of the present invention are described. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0068] 1) Drug molecule: The chemical structure of a drug, for example, a benzene ring is a drug molecule.

[0069] 2) Drug molecule properties: properties that characterize the structure of drug molecules. For example, drug molecules with polar groups have a strong affinity for water and can attract water molecules or dissolve in water. The surface of the solid material formed by such drug molecules is easily wetted by water, that is, the property of such drug molecules is hydrophilic. A certain type of drug molecules can cause direct or indirect damage after contacting with living organisms or entering living organisms, that is, the property of such drug molecules is biotoxicity (biohazard).

[0070] 3) Lead compound: referred to as lead compound, a compound with certain biological activity and chemical structure obtained through various pathways and means, used for further compound structure modification.

[0071] 4) Simplified Molecular Input Line Entry Specification (SMILES): A specification that uses the American Standard Code for Information Interchange (ASCII) string to clearly describe the molecular structure. SMILES expression can use a string of characters to describe a three-dimensional chemical structure. For example, the SMILES expression of cyclohexane (C6H12) is C1CCCCC1, that is, C1CCCCC1 represents cyclohexane.

[0072] 5) Syntax rule: The mapping relationship from one node to at least one node, that is, the conversion relationship between symbols, defines the partial structure of the drug molecule in the form of vector or index. A drug molecule has at least one grammatical rule, and at least one grammatical rule is composed into a grammatical rule set, that is, a grammatical rule sequence, which is used to form a tree for characterizing a drug molecule and uniquely identifying the drug molecule. In the field of artificial intelligence, the decoder can identify the grammatical rule to determine whether the grammatical rule is valid, but the decoder cannot identify the validity of the SMILES expression.

[0073] In order to at least solve the above-mentioned technical problems of the related art, the embodiments of the present invention provide a drug molecule processing method, device, electronic device and storage medium based on artificial intelligence, which can obtain drug molecules that meet the requirements of specific drug molecule properties and meet the user's needs for drug molecules. The following describes an exemplary application of an artificial intelligence drug molecule processing device provided by an embodiment of the present invention. The artificial intelligence drug molecule processing device provided by an embodiment of the present invention may be a server, such as a server deployed in the cloud, which provides users with effective drug molecules corresponding to the drug molecule attribute requirements based on drug molecule attribute requirements provided by other devices or users. For example, the server obtains request data including drug molecule attribute requirements based on other devices, extracts drug molecule attribute requirements based on drug molecule attribute requirement data in the request data, and obtains effective drug molecules that meet the drug molecule attribute requirements based on the drug molecule attribute requirements through a drug molecule model; it may also be various types of user terminals such as laptops, tablet computers, desktop computers, mobile devices (e.g., mobile phones, personal digital assistants), such as handheld terminals, which obtain effective drug molecules that meet the drug molecule attribute requirements based on the drug molecule attribute requirements input by the user on the handheld terminal, and display them on the display interface of the handheld terminal, or play the drug molecules that meet the drug molecule attribute requirements in the form of voice broadcast through a loudspeaker, so that users can quickly obtain the required drugs based on the drug molecules, avoiding the need to consult other people or professionals to obtain the required drugs, thereby improving the user experience.

[0074] As an example, see Figure 1 , Figure 1 It is a schematic diagram of an application scenario of an artificial intelligence-based drug molecule processing system 10 provided in an embodiment of the present invention. The terminal 200 is connected to the server 100 via a network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0075] The terminal 200 can be used to obtain the drug molecule property requirements. For example, when the user inputs the drug molecule property requirements through the input interface, after the input is completed, the terminal automatically obtains the drug molecule property requirements and generates a generation request for the drug molecule.

[0076] In some embodiments, the terminal 200 locally executes the artificial intelligence-based drug molecule processing method provided by the embodiment of the present invention to complete the input drug molecule property requirements and obtain valid drug molecules that meet the drug molecule property requirements. For example, an application (Application, APP) is installed on the terminal 200, such as a drug query assistant APP. The user inputs the drug molecule property requirements in the drug query assistant APP. The terminal 200 obtains valid drug molecules that meet the drug molecule property requirements based on the drug molecule property requirements, and displays the drug molecules on the display interface 210 of the terminal 200, or plays the drug molecules through voice broadcast, so that the user can quickly obtain the required drug based on the drug molecule.

[0077] In some embodiments, the terminal 200 may also send the drug molecule property requirements input by the user on the terminal 100 to the server 100 through the network 300, and call the artificial intelligence-based drug molecule processing function provided by the server 100. The server 100 obtains valid drug molecules that meet the drug molecule property requirements through the artificial intelligence-based drug molecule processing method provided in the embodiment of the present invention. For example, a drug query assistant APP is installed on the terminal 200, and the user enters the drug molecule property requirements in the drug query assistant APP. The terminal sends request data for the drug molecule property requirements to the server 100 through the network 300. After receiving the request data, the server 100 extracts the drug molecule property requirements in the request data, obtains valid drug molecules that meet the drug molecule property requirements based on the drug molecule property requirements, and returns the drug molecules to the drug query assistant APP, displays the drug molecules on the display interface 210 of the terminal 200, or plays the drug molecules by voice broadcast, or the server 100 directly uses the drug molecules so that the user can quickly obtain the required drugs based on the drug molecules.

[0078] The structure of the drug molecule processing device based on artificial intelligence provided by the embodiment of the present invention is further described. The drug molecule processing device based on artificial intelligence can be various terminals, such as mobile phones, computers, etc., or can be Figure 1 A server 100 is shown.

[0079] See also Figure 2 , Figure 2 is a schematic diagram of the structure of a drug molecule processing device 500 based on artificial intelligence provided by an embodiment of the present invention, Figure 2The artificial intelligence-based drug molecule processing device 500 shown 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 artificial intelligence-based drug molecule processing device 500 are coupled together through a bus system 540. It can be understood that the bus system 540 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 540 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, in Figure 2 Various buses are labeled as bus system 540 .

[0080] 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., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0081] The user interface 530 includes one or more output devices 531 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 530 also includes one or more input devices 532, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0082] The memory 550 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 550 described in the embodiment of the present invention is intended to include any suitable type of memory. The memory 550 optionally includes one or more storage devices that are physically far away from the processor 510.

[0083] In some embodiments, the memory 550 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0084] Operating system 551, including system programs for processing various basic system services and performing hardware-related tasks, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0085] A network communication module 552, for reaching other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 include: Bluetooth, Wireless Compatibility Certification (WiFi), and Universal Serial Bus (USB);

[0086] a display module 553 for enabling presentation of information via one or more output devices 531 (e.g., display screen, speaker, etc.) associated with the user interface 530 (e.g., a user interface for operating peripherals and displaying content and information);

[0087] The input processing module 554 is used to detect one or more user inputs or interactions from one of the one or more input devices 532 and translate the detected inputs or interactions.

[0088] In some embodiments, the artificial intelligence-based drug molecule processing device provided in the embodiments of the present invention can be implemented in a combination of software and hardware. As an example, the artificial intelligence-based drug molecule processing device provided in the embodiments of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the artificial intelligence-based drug molecule processing method provided in the embodiments of the present invention. For example, the processor in the form of a hardware decoding processor can adopt one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs) or other electronic components.

[0089] In other embodiments, the drug molecule processing device based on artificial intelligence provided by the embodiments of the present invention can be implemented in a software manner. Figure 2 An artificial intelligence-based drug molecule processing device 555 stored in a memory 550 is shown, which can be software in the form of a program and a plug-in, and includes a series of modules, including a drug molecule model and a first training module 5556 and a second training module 5558 for training the drug molecule model; wherein the drug molecule model is used to implement the function of obtaining effective drug molecules that meet the requirements of drug molecule properties in the artificial intelligence-based drug molecule processing method provided in an embodiment of the present invention, including a series of modules (to be explained below), and the first training module 5556 and the second training module 5558 are used to implement the training function of the drug molecule model provided in an embodiment of the present invention.

[0090] In conjunction with the exemplary application and implementation of the server provided in the embodiment of the present invention, the artificial intelligence-based drug molecule processing method provided in the embodiment of the present invention is explained. According to the above, it can be understood that the artificial intelligence-based drug molecule processing method provided in the embodiment of the present invention can be used by various types of artificial intelligence-based drug molecule processing devices, such as smart terminals and servers.

[0091] See also Figure 3 and Figure 4A , Figure 3 FIG. 5 is a schematic diagram of the structure of the drug molecule processing device 555 based on artificial intelligence provided by an embodiment of the present invention, showing the drug molecule processing process based on artificial intelligence. Figure 4A The flowchart of the drug molecule processing method based on artificial intelligence provided by the embodiment of the present invention realizes the training function of the coding model in the drug molecule model. The drug molecule processing method based on artificial intelligence provided by the embodiment of the present invention is applied to the server as an example. Figure 3 right Figure 4A The steps shown are explained.

[0092] In step 101A, the grammatical rules of the training samples are sampled by using the mean vector and the covariance matrix in the encoder in the coding model to obtain the feature vector of the training samples.

[0093] The drug molecule processing method based on artificial intelligence is applied to the drug molecule model, wherein the drug molecule model includes a molecular property prediction model and an encoding model. The training process of the drug molecule model is realized by training the molecular property prediction model and the encoding model respectively.

[0094] In order to implement the training process of the encoding model in the drug molecule model, the training samples can be input into the encoding model. The encoding model includes at least one encoder and at least one decoder, wherein the encoder and the decoder are in a one-to-one correspondence relationship.

[0095] In order to train the encoding model so that it can output the grammatical rules corresponding to drug molecules, the SEILES expression of the training sample (molecular structure in the public data set) can be converted into a series of grammatical rules, i.e., a grammatical rule sequence, based on a pre-defined grammatical rule set. Figure 6 As shown, the grammatical rules of the training samples are input into the encoder in the encoding model. The encoder can output the mean vector and covariance matrix in the multivariate Gaussian distribution that conforms to the grammatical rules according to the input grammatical rules. The grammatical rules of the training samples are sampled according to the mean vector and covariance matrix to obtain the feature vector of the training samples.

[0096] As an example, see Figure 3 The first training module 5556 samples the grammatical rules of the training samples through the mean vector and covariance matrix in the encoder in the encoding model to obtain the feature vector of the training samples.

[0097] In some embodiments, before sampling the grammatical rules of the training samples through the mean vector and covariance matrix in the encoder in the encoding model to obtain the feature vector of the training samples, it also includes: disassembling the grammatical rule sequence corresponding to the molecular linear input specification of the training samples to obtain the grammatical rule of at least one training sample.

[0098] Since the coding model includes at least one encoder and at least one decoder, where the encoder and the decoder are in a one-to-one correspondence. Figure 6 As shown, the SEILES expression of the training sample can be converted into a grammatical rule sequence for splitting to obtain the grammatical rule of at least one training sample, and the grammatical rule of the at least one training sample can be input into the corresponding encoders in the encoding model respectively.

[0099] As an example, see Figure 3 The disassembly module 5557 disassembles the grammatical rule sequence corresponding to the molecular linear input specification of the training sample to obtain the grammatical rule of at least one training sample, so that the subsequent first training module 5556 can train the encoding model.

[0100] In step 102A, the feature vector of the training sample is decoded by a decoder in the coding model to obtain a second grammar rule corresponding to the training sample.

[0101] After sampling the grammatical rules of the training sample through the mean vector and covariance matrix in the encoder to obtain the feature vector of the training sample, the feature vector of the training sample will be input into the decoder in the corresponding encoding model. The corresponding decoder decodes the feature vector of the training sample to obtain the second grammatical rule of the corresponding training sample.

[0102] In some embodiments, where the decoder is the nth decoder, the feature vector of the training sample is decoded by the decoder in the coding model to obtain a second grammar rule corresponding to the training sample, including: using the nth decoder in the coding model, the output of the n-1th encoder and the feature vector of the training sample are decoded to obtain the second grammar rule corresponding to the training sample; where n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the coding model.

[0103] like Figure 6As shown, after the nth grammatical rule in the training sample is input to the nth encoder, the feature vector of the nth grammatical rule in the training sample is obtained by sampling, and the feature vector of the nth grammatical rule in the training sample is input to the nth encoder, and the output result output by the n-1th encoder is also input to the nth decoder. The nth decoder decodes the output result of the n-1th encoder and the feature vector of the nth grammatical rule in the training sample to obtain the second grammatical rule corresponding to the training sample (the grammatical rule output by the training sample through the encoding model). Since the output result of the n-1th encoder is added to the nth encoder, a direct dependency relationship between the grammatical rules can be established, and through the dependency relationship, the encoder can effectively identify the wrong grammatical rule, that is, the probability of the second grammatical rule to be generated is set to 0.

[0104] As an example, see Figure 3 The first training module 5556 decodes the feature vector of the training sample through the decoder in the encoding model to obtain the second grammar rule corresponding to the training sample.

[0105] In step 103A, model parameters of the encoding model are updated according to the grammatical rules of the training samples and the second grammatical rules of the training samples.

[0106] After obtaining the second grammatical rule corresponding to the training sample through the decoder in the coding model, the value of the loss function of the coding model can be obtained according to the grammatical rule of the training sample and the second grammatical rule of the training sample, and it is determined whether the value of the loss function exceeds a preset threshold. When the value of the loss function exceeds the preset threshold, the error signal of the coding model is determined based on the loss function, the error information is back-propagated in the coding model, and the model parameters of each layer are updated during the propagation process.

[0107] As an example, see Figure 3 The first training module 5556 updates the model parameters of the encoding model according to the grammatical rules of the training samples and the second grammatical rules of the training samples.

[0108] Here, back propagation is explained. The training sample data is input into the input layer of the neural network model, passes through the hidden layer, and finally reaches the output layer and outputs the result. This is the forward propagation process of the neural network model. Since there is an error between the output result of the neural network model and the actual result, the error between the output result and the actual value is calculated, and the error is backpropagated from the output layer to the hidden layer until it propagates to the input layer. In the process of back propagation, the value of the model parameter is adjusted according to the error; the above process is continuously iterated until convergence. Among them, the encoding model belongs to the neural network model.

[0109] See also Figure 3 and Figure 4B , Figure 4B The flowchart of the drug molecule processing method based on artificial intelligence provided by the embodiment of the present invention is as follows: in step 101B, the drug molecule vector of the training sample is predicted by the molecular property prediction model to obtain the drug properties of the corresponding training sample; in step 102B, the value of the loss function of the molecular property prediction model is determined according to the drug properties of the training sample and the standard drug properties; in step 103B, the model parameters of the molecular property prediction model are updated according to the value of the loss function. Among them, Figure 4A Steps 101A-103A are shown as a training process for the encoding model in the drug molecule model. Figure 4B Steps 101B-103B are shown as a training process for the molecular property prediction model in the drug molecule model. There is no obvious sequence between steps 101A-103A and 101B-103B.

[0110] In some embodiments, after the drug molecule vector of the training sample is input into the molecular property prediction model, the molecular property prediction model will make predictions based on the drug molecule vector of the training sample and output the drug properties of the corresponding training sample, wherein the drug properties of the training sample may be the probability values ​​of the drug properties of the corresponding training sample, for example, the probability that the drug properties of the training sample are hydrophilic is 90%; the drug properties of the training sample may be the numerical values ​​of the drug properties of the corresponding training sample, for example, the drug properties (molar mass) of the training sample is 58.5 g / mol.

[0111] After obtaining the drug properties of the corresponding training samples, the value of the loss function of the molecular property prediction model can be determined according to the drug properties of the training samples and the drug properties of the standard, and it can be determined whether the value of the loss function exceeds the preset threshold. When the value of the loss function exceeds the preset threshold, the error signal of the molecular property prediction model is determined based on the loss function, the error information is back-propagated in the encoding model, and the model parameters of each layer are updated during the propagation process. When the drug property of the training sample can be the probability value of the drug property of the corresponding training sample, the standard drug property is the probability value of the drug property of the corresponding standard; when the drug property of the training sample can be the numerical value of the drug property of the corresponding training sample, the standard drug property is the numerical value of the drug property of the corresponding standard.

[0112] As an example, see Figure 3 The second training module 5558 predicts the drug molecule vector of the training sample through the molecular property prediction model to obtain the drug properties of the corresponding training sample, and determines the value of the loss function of the molecular property prediction model according to the drug properties of the training sample and the standard drug properties. Finally, according to the value of the loss function, the model parameters of the molecular property prediction model are updated.

[0113] Here, back propagation is explained. The training sample data is input into the input layer of the neural network model, passes through the hidden layer, and finally reaches the output layer and outputs the result. This is the forward propagation process of the neural network model. Since there is an error between the output result of the neural network model and the actual result, the error between the output result and the actual value is calculated, and the error is backpropagated from the output layer to the hidden layer until it propagates to the input layer. During the back propagation process, the value of the model parameter is adjusted according to the error; the above process is continuously iterated until convergence. Among them, the molecular property prediction model belongs to the neural network model.

[0114] See below Figure 3 and Figure 4C , Figure 4A : is a flow chart of a drug molecule processing method based on artificial intelligence provided by an embodiment of the present invention. The drug molecule processing method based on artificial intelligence is applied to a drug molecule model. The drug molecule model includes a molecular property prediction model and a coding model. The trained drug molecule model is used to obtain effective drug molecules that meet the drug molecule property requirements according to the drug molecule property requirements. Taking the drug molecule processing method based on artificial intelligence provided by an embodiment of the present invention applied to a server as an example, combined with the following Figure 3 right Figure 4C The steps shown are explained.

[0115] In step 201 , in response to a request for generating a drug molecule, a drug molecule property requirement is determined.

[0116] Here, the user can input the drug molecule property requirements, such as hydrophilicity, in the input interface of the terminal. When the input is completed, the terminal automatically generates a generation request for the drug molecule and sends the generation request for the drug molecule to the server. When the server receives the generation request for the drug molecule, it parses the generation request to obtain the drug molecule property requirements.

[0117] As an example, see Figure 3 , the first determination module 5551 in the drug molecule model responds to the generation request for the drug molecule, determines the drug molecule property requirements, and inputs the drug molecule property requirements into the prediction module 5553.

[0118] In step 202, a randomly sampled drug molecule vector is obtained.

[0119] In some embodiments, obtaining a randomly sampled drug molecule vector includes: performing random sampling processing on a set of drug molecule vectors that obey a multivariate Gaussian distribution to obtain a randomly sampled drug molecule vector.

[0120] Here, the drug molecule vector set (the set of all possible molecular structures) contains M drug molecule vectors, where M is a natural number greater than 1. Fig.10 As shown, the drug molecule vector set obeys the multivariate Gaussian distribution, the mean vector of the multivariate Gaussian distribution is set to 0, the covariance matrix is ​​set to the unit matrix, and the drug molecule vector set obeying the multivariate Gaussian distribution is randomly sampled to obtain randomly sampled drug molecule vectors, so as to facilitate subsequent prediction and transformation of the randomly sampled drug molecule vectors.

[0121] As an example, see Figure 3 , the randomly sampled drug molecule vector is obtained through the acquisition module 5552 in the drug molecule model, and the randomly sampled drug molecule vector is input into the prediction module 5553.

[0122] In step 203, the randomly sampled drug molecule vectors are transformed to obtain drug molecule vectors that meet the requirements of drug molecule properties.

[0123] After the server obtains the randomly sampled drug molecule vector, it can also transform the randomly sampled drug molecule vector through the trained molecular property prediction model to obtain a drug molecule vector that meets the drug molecule property requirements, thereby meeting the user's requirements for drug properties.

[0124] In some embodiments, a randomly sampled drug molecule vector is transformed to obtain a drug molecule vector that meets the drug molecule property requirements, including: predicting the randomly sampled drug molecule vector through a molecular property prediction model to obtain the drug properties corresponding to the randomly sampled drug molecule vector; comparing the drug properties corresponding to the randomly sampled drug molecule vector with the drug molecule property requirements, and updating the randomly sampled drug molecule vector according to the comparison result to obtain the drug molecule vector that meets the drug molecule property requirements.

[0125] After the server obtains the randomly sampled drug molecule vector, the randomly sampled drug molecule vector can first be predicted by the trained molecular property prediction model to determine the drug property corresponding to the randomly sampled drug molecule vector. The drug property can be a numerical value, for example, the drug property is molar mass, and the drug property corresponding to the randomly sampled drug molecule vector is compared with the drug molecule property requirement to obtain a comparison result, wherein the drug molecule property requirement can also be a numerical value. According to the comparison result, if it is determined that the drug property corresponding to the randomly sampled drug molecule vector is different from the drug molecule property requirement, the randomly sampled drug molecule vector is updated to obtain a drug molecule vector that meets the drug molecule property requirement.

[0126] In some embodiments, a randomly sampled drug molecule vector is transformed to obtain a drug molecule vector that meets the drug molecule property requirements, including: predicting the randomly sampled drug molecule vector through a molecular property prediction model to obtain a probability value of at least one drug property corresponding to the randomly sampled drug molecule vector; subtracting the probability value of at least one drug property corresponding to the randomly sampled drug molecule vector from the probability value of the drug property in the drug molecule property requirements to obtain a corresponding difference; when the difference is greater than a difference threshold, updating the randomly sampled drug molecule vector according to the drug molecule property requirements to obtain a drug molecule vector that meets the drug molecule property requirements.

[0127] After the server obtains the randomly sampled drug molecule vector, the randomly sampled drug molecule vector can first be predicted by the trained molecular property prediction model to obtain the probability value of at least one drug property corresponding to the randomly sampled drug molecule vector, for example, the probability value of hydrophilicity corresponding to the randomly sampled drug molecule vector is 90%, and the biological toxicity is 50%. The probability value of at least one drug property corresponding to the randomly sampled drug molecule vector is subtracted from the probability value of the drug property in the drug molecule property requirement to obtain the corresponding difference value, for example, if the probability value of hydrophilicity corresponding to the randomly sampled drug molecule vector is 90%, and the probability value of hydrophilicity in the drug molecule property requirement is 85%, then the corresponding difference value is 5%. When the difference is greater than the difference threshold, the randomly sampled drug molecule vector is updated according to the drug molecule attribute requirements to obtain the drug molecule vector that meets the drug molecule attribute requirements. For example, if the difference threshold is 6%, the corresponding difference of 5% is less than the difference threshold, then the randomly sampled drug molecule vector is updated according to the drug molecule attribute requirements to obtain the drug molecule vector that meets the drug molecule attribute requirements. When the difference is less than or equal to the difference threshold, the randomly sampled drug molecule vector meets the drug molecule attribute requirements and there is no need to update the randomly sampled drug molecule vector.

[0128] As an example, see Figure 3 , the prediction module 5553 in the drug molecule model transforms the randomly sampled drug molecule vector through the molecular property prediction model to obtain the drug molecule vector that meets the drug molecule property requirements, and the drug molecule vector that meets the drug molecule property requirements is input into the encoding module 5554.

[0129] In step 204, the drug molecule vector that meets the drug molecule attribute requirements is decoded to obtain a first grammatical rule corresponding to the drug molecule vector.

[0130] When the server obtains the drug molecule vector that meets the drug molecule attribute requirements, it can also decode the drug molecule vector that meets the drug molecule attribute requirements through the decoder in the trained encoding model to obtain the first grammatical rule corresponding to the drug molecule vector, thereby introducing the constraints of the grammatical rules in the encoding model to obtain the first grammatical rule corresponding to the drug molecule vector, and the drug molecules subsequently obtained according to the grammatical rules are more effective.

[0131] In some embodiments, a drug molecule vector that meets the drug molecule property requirements is decoded to obtain a first grammatical rule corresponding to the drug molecule vector, including: decoding the drug molecule vector that meets the drug molecule property requirements through at least one decoder of the encoding model to generate at least one first grammatical rule for the drug molecule vector; the encoding model includes multiple decoders.

[0132] In the process of training the coding model, the coding model has N encoders and N decoders corresponding to the encoders. When the coding model is trained, the drug molecule vector that meets the requirements of the drug molecule properties can be decoded by the N decoders to obtain the first grammatical rule corresponding to the drug molecule vector.

[0133] After the server obtains the drug molecule vector that meets the drug molecule property requirements, the drug molecule vector that meets the drug molecule property requirements is input into N encoders respectively. The drug molecule vector that meets the drug molecule property requirements is decoded by at least one decoder of the encoding model to generate at least one first grammar rule for the drug molecule vector.

[0134] In some embodiments, a drug molecule vector that meets the drug molecule property requirements is decoded by at least one decoder of the coding model to generate at least one first grammatical rule for the drug molecule vector, including: decoding the output of the n-1th decoder and the drug molecule vector that meets the drug molecule property requirements by the nth decoder in the coding model to obtain the first grammatical rule of the drug molecule vector corresponding to the nth decoder; wherein n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the coding model.

[0135] like Fig.10As shown, after the drug molecule vectors that meet the drug molecule property requirements are input into N encoders respectively, the n-1th decoder outputs the corresponding output result to the nth decoder in the encoding model according to the drug molecule vector that meets the drug molecule property requirements, and the nth decoder decodes the output of the n-1th decoder and the drug molecule vector that meets the drug molecule property requirements to obtain the first grammatical rule of the drug molecule vector corresponding to the nth decoder, and inputs the first grammatical rule of the drug molecule vector corresponding to the nth decoder to the n+1th decoder until N decoders output corresponding output results.

[0136] As an example, see Figure 3 , the encoding module 5553 in the drug molecule model decodes the drug molecule vector that meets the drug molecule property requirements through the decoder in the encoding model to obtain the first grammatical rule corresponding to the drug molecule vector, and inputs the first grammatical rule corresponding to the drug molecule vector into the second determination module 5555.

[0137] In step 205, according to the mapping relationship between the grammatical rules and the drug molecules, the drug molecules corresponding to the first grammatical rule are generated to respond to the generation request.

[0138] After the server obtains the first grammatical rule, it can also obtain the mapping relationship between the grammatical rule and the drug molecule pre-set by the user locally, so as to determine the drug molecule corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecule, in response to the generation request. In addition, the mapping relationship between the grammatical rule and the drug molecule can also be set on other electronic devices. When the server needs to obtain the mapping relationship between the grammatical rule and the drug molecule, the server can obtain the mapping relationship between the grammatical rule and the drug molecule by interacting with other electronic devices, so as to determine the drug molecule corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecule, in response to the generation request.

[0139] In some embodiments, a drug molecule corresponding to a first grammatical rule is generated according to a mapping relationship between grammatical rules and drug molecules, including: combining at least one first grammatical rule to obtain a grammatical rule sequence corresponding to a drug molecule vector; and obtaining a drug molecule corresponding to the first grammatical rule according to a mapping relationship between the grammatical rule sequence and the drug molecules and the grammatical rule sequence corresponding to the drug molecule vector.

[0140] After the decoder in the coding model outputs the corresponding first grammatical rules in sequence, at least one of the output first grammatical rules is combined to obtain a grammatical rule sequence corresponding to the drug molecule vector. After the server obtains the grammatical rule sequence, it can also obtain a mapping relationship between the grammatical rule sequence and the drug molecules, thereby determining the corresponding drug molecules based on the mapping relationship between the grammatical rule sequence and the drug molecules. In addition, the mapping relationship between the grammatical rule sequence and the drug molecules can also be set on other electronic devices. When the server needs to obtain the mapping relationship between the grammatical rule sequence and the drug molecules, the server can obtain the mapping relationship between the grammatical rule sequence and the drug molecules by interacting with other electronic devices, thereby determining the drug molecules corresponding to the first grammatical rule based on the mapping relationship between the grammatical rule sequence and the drug molecules.

[0141] In some embodiments, the drug molecule corresponding to the first grammar rule is determined according to the mapping relationship between the grammar rules and the drug molecules, including: combining at least one first grammar rule to obtain a grammar rule sequence corresponding to the drug molecule vector; obtaining a molecular linear input specification corresponding to the drug molecule vector according to the mapping relationship between the molecular linear input specification and the grammar rule sequence, and the grammar rule sequence corresponding to the drug molecule vector; obtaining the drug molecule corresponding to the first grammar rule according to the mapping relationship between the molecular linear input specification and the drug molecule, and the molecular linear input specification corresponding to the drug molecule vector.

[0142] After the decoder in the coding model outputs the corresponding first grammar rule in sequence, at least one of the output first grammar rules is combined to obtain a grammar rule sequence corresponding to the drug molecule vector. After the server obtains the grammar rule sequence, it can also obtain a mapping relationship between the molecular linear input specification and the grammar rule sequence, so as to determine the molecular linear input specification of the corresponding drug molecule vector according to the mapping relationship between the molecular linear input specification and the grammar rule sequence. The server can also interact with other electronic devices (with a mapping relationship between the molecular linear input specification and the grammar rule sequence) to obtain a mapping relationship between the molecular linear input specification and the grammar rule sequence, so as to determine the molecular linear input specification of the corresponding drug molecule vector according to the mapping relationship between the molecular linear input specification and the grammar rule sequence. When the molecular linear input specification of the corresponding drug molecule vector is determined, the mapping relationship between the molecular linear input specification and the drug molecule on the local or other device can be obtained, and the drug molecule corresponding to the first grammar rule is obtained according to the mapping relationship between the molecular linear input specification and the drug molecule and the molecular linear input specification of the corresponding drug molecule vector.

[0143] As an example, see Figure 3, through the second determination module 5555 in the drug molecule model, according to the mapping relationship between the grammatical rules and the drug molecules, the drug molecules corresponding to the first grammatical rule can be determined to respond to the generation request.

[0144] So far, we have combined the artificial intelligence-based drug molecule processing method provided in the embodiment of the present invention and its exemplary application in the artificial intelligence-based drug molecule processing device. The following will continue to explain the solution of artificial intelligence-based drug molecule processing implemented by the cooperation of various modules in the artificial intelligence-based drug molecule processing device 555 provided in the embodiment of the present invention.

[0145] A first determination module 5551 is used to determine the drug molecule property requirements in response to a generation request for a drug molecule;

[0146] An acquisition module 5552 is used to acquire randomly sampled drug molecule vectors;

[0147] The prediction module 5553 is used to transform the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements;

[0148] The encoding module 5554 is used to decode the drug molecule vector that meets the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector;

[0149] The second determination module 5555 is used to generate the drug molecule corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecule, so as to respond to the generation request.

[0150] In some embodiments, the acquisition unit 5552 is further used to perform random sampling processing on a set of drug molecule vectors that obey a multivariate Gaussian distribution to obtain randomly sampled drug molecule vectors.

[0151] In some embodiments, the prediction module 5553 is also used to predict the randomly sampled drug molecule vector through a molecular property prediction model to obtain the drug properties corresponding to the randomly sampled drug molecule vector; compare the drug properties corresponding to the randomly sampled drug molecule vector with the drug molecule property requirements, and update the randomly sampled drug molecule vector according to the comparison result to obtain a drug molecule vector that meets the drug molecule property requirements.

[0152] In some embodiments, the prediction module 5553 is also used to predict the randomly sampled drug molecule vector through a molecular property prediction model to obtain a probability value of at least one drug property corresponding to the randomly sampled drug molecule vector; subtract the probability value of at least one drug property corresponding to the randomly sampled drug molecule vector from the probability value of the drug property in the drug molecule property requirement to obtain a corresponding difference; when the difference is greater than a difference threshold, update the randomly sampled drug molecule vector according to the drug molecule property requirement to obtain a drug molecule vector that meets the drug molecule property requirement.

[0153] In some embodiments, the encoding model 5554 includes multiple decoders; the encoding module is also used to decode the drug molecule vector that meets the drug molecule property requirements through at least one decoder of the encoding model to generate at least one first grammar rule for the drug molecule vector.

[0154] In some embodiments, the encoding module 5554 is also used to decode the output of the n-1th decoder and the drug molecule vector that meets the drug molecule property requirements through the nth decoder in the encoding model to obtain the first grammatical rule of the drug molecule vector corresponding to the nth decoder; wherein n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the encoding model.

[0155] In some embodiments, the second determination module 5555 is further used to combine at least one first grammatical rule to obtain a grammatical rule sequence corresponding to the drug molecule vector; based on the mapping relationship between the grammatical rule sequence and the drug molecule, and the grammatical rule sequence corresponding to the drug molecule vector, the drug molecule corresponding to the first grammatical rule is obtained.

[0156] In some embodiments, the second determination module 5555 is also used to combine at least one first grammar rule to obtain a grammar rule sequence corresponding to the drug molecule vector; obtain the molecular linear input specification corresponding to the drug molecule vector according to the mapping relationship between the molecular linear input specification and the grammar rule sequence, and the grammar rule sequence corresponding to the drug molecule vector; obtain the drug molecule corresponding to the first grammar rule according to the mapping relationship between the molecular linear input specification and the drug molecule, and the molecular linear input specification corresponding to the drug molecule vector.

[0157] In some embodiments, the artificial intelligence-based drug molecule processing device 555 further includes:

[0158] The first training module 5556 is used to sample the grammatical rules of the training samples through the mean vector and covariance matrix in the encoder in the coding model to obtain the feature vector of the training samples; decode the feature vector of the training samples through the decoder in the coding model to obtain the second grammatical rules corresponding to the training samples; and update the model parameters of the coding model according to the grammatical rules of the training samples and the second grammatical rules of the training samples.

[0159] In some embodiments, the artificial intelligence-based drug molecule processing device 555 further includes:

[0160] The disassembly module 5557 is used to disassemble the grammatical rule sequence corresponding to the molecular linear input specification of the training sample to obtain the grammatical rule of at least one training sample.

[0161] In some embodiments, the decoder is the nth decoder, and the first training module 5556 is also used to decode the output of the n-1th encoder and the feature vector of the training sample through the nth decoder in the encoding model to obtain a second grammar rule corresponding to the training sample; wherein n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the encoding model.

[0162] In some embodiments, the artificial intelligence-based drug molecule processing device 555 further includes:

[0163] The second training module 5558 is used to predict the drug molecule vector of the training sample through the molecular property prediction model to obtain the drug properties corresponding to the training sample; determine the value of the loss function of the molecular property prediction model according to the drug properties of the training sample and the standard drug properties; and update the model parameters of the molecular property prediction model according to the value of the loss function.

[0164] The embodiment of the present invention further provides a storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the drug molecule processing method based on artificial intelligence provided by the embodiment of the present invention, for example, Figures 4A-4C An artificial intelligence-based drug molecule processing method is shown.

[0165] In some embodiments, the storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EE PROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

[0166] In some embodiments, executable instructions may be in 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 stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0167] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (for example, files storing one or more modules, subroutines, or code portions).

[0168] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0169] Below, an exemplary application of an embodiment of the present invention in an actual application scenario of artificial intelligence-based drug molecule processing will be described.

[0170] In the related art, there are two main methods for generating candidate drug molecules based on machine learning, namely, two types of models based on variational auto-encoders (VAE) and generative adversarial networks (GENER ativeAdversarial Networks). In the scheme based on variational auto-encoders, the drug molecule structure is often represented as a SMILES expression, thereby converting the problem of generating a drug molecule structure into the problem of generating a sequence.

[0171] Based on the variational autoencoder framework, a network structure containing three layers of gated recurrent units (GRU) is used in the encoding and decoding parts for feature extraction and generation of sequence data. In addition, the variational autoencoder can be encoded and decoded through a multi-layer convolutional neural network. At the same time, grammatical rule constraints are introduced to improve the effectiveness of drug molecule generation. The grammatical rules can be modeled into a tree structure to ensure that the generated drug molecule structure can fully meet the requirements of chemical laws. By adopting the Bayesian optimization method, drug molecule structures that meet specific attribute requirements can be generated.

[0172] In the technical solution based on variational autoencoders, the constraints of grammatical rules are not fully utilized, resulting in a large number of invalid results (invalid molecules) in the generated molecular structures. Even if the constraints of grammatical rules are taken into account, when generating molecular structures that meet specific property requirements, a Bayesian optimization method (Gaussian Processes) is used. However, when the number of samples is large, this method has the problem of slow training speed, that is, the time complexity is the cube of the number of samples, and the training process is not efficient enough.

[0173] In the embodiments of the present invention, on the one hand, the effectiveness of generating drug molecules is improved by introducing constraints of grammatical rules, and on the other hand, the problem of slow training speed is avoided by integrating molecular property prediction models, so that drug molecular structures that meet specific property requirements can be generated more efficiently.

[0174] like Figure 5 As shown in Figure 2, in the process of drug development, after completing target identification (Target Selection), a large number of candidate drug molecules (Hit Generation) need to be obtained, that is, candidate molecule generation, which is about 10 6 molecules; 10 6 After the subsequent lead compound identification screening, there are about 10 4 molecules for subsequent lead compound optimization; 10 4 After the molecules are screened and modified through lead optimization, the most promising molecular structure for drug development is obtained. 3 molecules; 10 3 The molecules finally enter the subsequent animal tests and clinical trials for final verification, and about 1 molecule is obtained. Therefore, in the stage of candidate drug molecule generation, if a large number of effective molecular structures that meet specific property requirements can be efficiently generated, the development efficiency of the drug development process can be improved and the time expenditure and experimental cost of drug development can be reduced.

[0175] The embodiment of the present invention aims at the problem of generating candidate drug molecules in the drug development process, and proposes an artificial intelligence-based drug molecule processing solution. By combining grammatical rule constraints and molecular property prediction models, a large number of effective molecular structures that meet specific property requirements can be efficiently generated for subsequent drug screening and testing.

[0176] The implementation process of the embodiment of the present invention is divided into three stages, namely: 1) training of the variational autoencoder model; 2) training of the molecular property prediction model; 3) generating drug molecules based on the variational autoencoder and the molecular property prediction model.

[0177] In the first stage, a variational autoencoder model (encoding model) is trained based on a large amount of molecular structure data for subsequent drug molecule generation. Figure 6 As shown in the figure, two recurrent neural networks (Recurrent Neural Networks) containing multiple layers of GRU units are used as the encoder and decoder of the variational autoencoder respectively. At the same time, the output result of the previous decoder is added to the input of the next decoder, so as to more directly model the front-end dependency relationship of the sequence data.

[0178] In terms of training samples, the molecular structures provided in the public data set ZINC can be used, and based on a pre-defined set of grammatical rules, the SMILES expression of each molecule can be converted into a series of grammatical rules, that is, a grammatical rule sequence, and the grammatical rules can be used as the input of the variational autoencoder. Among them, the output of the variational autoencoder is also a series of grammatical rules, corresponding to the SMILES expression of the generated molecular structure. For any SMILES expression, based on the grammatical rule set, it can be parsed into a corresponding tree structure, and the tree structure can be traversed in pre-order to obtain the corresponding grammatical rule sequence. Similarly, any legal grammatical rule sequence can construct a corresponding tree structure, and then obtain the SMILES expression of the corresponding molecular structure. For example Figure 7-8 As shown, Figure 7 is a schematic diagram of a molecular linear input specification grammar rule set provided by an embodiment of the present invention, Figure 8 is a schematic diagram of the conversion between the SMILES expression and the grammatical rule sequence provided by the embodiment of the present invention. The SMILES expression of the benzene ring is c1cccc c1. Figure 7 The grammar rule set shown can be parsed into the corresponding Figure 8 The tree structure shown in Figure 1 can be obtained by traversing the tree structure in pre-order. Figure 8 The sequence of grammar rules shown.

[0179] In the embodiment of the present invention, based on the variational autoencoder model obtained by the first stage training, for any drug molecular structure, its characteristic expression (feature vector) in the latent space can be obtained. In the second stage, according to the characteristic expression of each drug molecular structure in the latent space and the molecular attribute value of the drug molecule (the output of the molecular attribute prediction model), a multi-layer perceptron model is trained to predict the molecular attribute value, such as Fig. 9 As shown, for example, the properties of drug molecules output by the multilayer perceptron can be predicted as to whether they are hydrophilic or biotoxic.

[0180] In the embodiment of the present invention, the third stage: Fig.10 As shown, based on the variational autoencoder model obtained by the first stage training and the molecular property prediction model obtained by the second stage training, in the process of using the variational autoencoder model to generate molecular structures, the feature expression in the latent space obtained by random sampling is first optimized based on the molecular property prediction model to meet the specific property requirements, and then input into the decoder of the variational autoencoder, so that the obtained drug molecular structure can meet the specific molecular property requirements while ensuring the structural validity. Among them, according to the multi-layer perceptron model, the molecular property prediction result corresponding to the feature expression in the current latent space can be calculated, and the difference with the expected molecular property requirement can be calculated as the loss function of the molecular property prediction model, and then the input feature expression is gradient updated to reduce the difference between the molecular property prediction value and the expected value.

[0181] In the embodiment of the present invention, on the one hand, by introducing the constraints of grammatical rules, the efficiency of molecule generation is improved, and on the other hand, by integrating the molecular property prediction model, the molecular structure that meets the specific property requirements can be generated more efficiently. Compared with not integrating the molecular property prediction model, or using other molecular structure generation models (rather than the variational autoencoder based on grammatical rule constraints), given the same drug molecule property requirements, the drug molecule model in the embodiment of the present invention can obtain a better drug molecule structure and improve the effectiveness of generating drug molecules. Fig.11 As shown, "GVAE-Design" is the final experimental effect of the model of the embodiment of the present invention, "GVAE-Random" is the experimental effect when the molecular property prediction model is not integrated, "JTVAE-Random" is the experimental effect of the JTVAE model (another molecular structure generation model, not based on grammatical rule constraints) when the molecular property prediction model is not integrated, and "JTVAE-Design" is the experimental effect of the JTVAE model when the molecular property prediction model is integrated. Fig.11 It can be seen that the model in the embodiment of the present invention has the highest efficiency in the five molecular attribute values ​​(logP, SA, NPQ, QED, weight), which are basically above 99%. Only the quantitative index of drug molecules (Quantitative Estimation of Drug-likeness, QED) is slightly lower, which is 96.8%.

[0182] In summary, through the variational autoencoder model obtained through the first stage training and the molecular property prediction model obtained through the second stage training in the embodiment of the present invention, drug molecules can be obtained, which has the following beneficial effects:

[0183] 1. Improve the effectiveness of generating drug molecules by introducing constraints of grammatical rules.

[0184] 2. By integrating the molecular property prediction model, the problem of slow training speed is avoided, so that drug molecular structures that meet specific property requirements can be generated more efficiently.

[0185] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.

Claims

1. A drug molecule processing method based on artificial intelligence, characterized in that: The method comprises: In response to a request for generating a drug molecule, determining a drug molecule property requirement; Obtaining a randomly sampled drug molecule vector, wherein the randomly sampled drug molecule vector is obtained by performing a random sampling process on a set of drug molecule vectors that obey a multivariate Gaussian distribution; Transforming the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements; Decoding the drug molecule vector that meets the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector, wherein the first grammatical rule is used to represent a partial structure of the drug molecule in a vector or index manner; According to the mapping relationship between the grammatical rules and the drug molecules, the drug molecules corresponding to the first grammatical rules are generated to respond to the generation request.

2. The method according to claim 1, characterized in that The transforming process of the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements includes: Predicting the randomly sampled drug molecule vector using a molecular property prediction model to obtain drug properties corresponding to the randomly sampled drug molecule vector; The drug property corresponding to the randomly sampled drug molecule vector is compared with the drug molecule property requirement, and the randomly sampled drug molecule vector is updated according to the comparison result to obtain a drug molecule vector that meets the drug molecule property requirement.

3. The method according to claim 1, characterized in that The transforming process of the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements includes: Predicting the randomly sampled drug molecule vector using a molecular property prediction model to obtain a probability value of at least one drug property corresponding to the randomly sampled drug molecule vector; Subtracting the probability value of at least one drug attribute corresponding to the randomly sampled drug molecule vector from the probability value of the drug attribute in the drug molecule attribute requirement to obtain a corresponding difference value; When the difference is greater than a difference threshold, the randomly sampled drug molecule vector is updated according to the drug molecule property requirement to obtain a drug molecule vector that meets the drug molecule property requirement.

4. The method according to claim 1, characterized in that: The decoding of the drug molecule vector that meets the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector includes: Decoding the drug molecule vector that meets the drug molecule property requirements through at least one decoder of the coding model to generate at least one first grammar rule for the drug molecule vector; The coding model includes multiple decoders.

5. The method according to claim 4, characterized in that The method of decoding the drug molecule vector that meets the drug molecule property requirements by at least one decoder of the coding model to generate at least one first grammar rule for the drug molecule vector includes: Decoding the output of the n-1th decoder and the drug molecule vector that meets the drug molecule property requirements through the nth decoder in the encoding model to obtain a first grammatical rule of the drug molecule vector corresponding to the nth decoder; Wherein, n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the coding model.

6. The method according to claim 1, characterized in that The step of generating drug molecules corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecules includes: Combining at least one first grammatical rule to obtain a grammatical rule sequence corresponding to the drug molecule vector; The drug molecule corresponding to the first grammatical rule is obtained according to the mapping relationship between the grammatical rule sequence and the drug molecule, and the grammatical rule sequence corresponding to the drug molecule vector.

7. The method according to claim 1, characterized in that The step of generating drug molecules corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecules includes: Combining at least one first grammatical rule to obtain a grammatical rule sequence corresponding to the drug molecule vector; According to the mapping relationship between the molecular linear input specification and the grammatical rule sequence, and the grammatical rule sequence corresponding to the drug molecular vector, the molecular linear input specification corresponding to the drug molecular vector is obtained; The drug molecule corresponding to the first grammar rule is obtained according to the mapping relationship between the molecular linear input specification and the drug molecule, and the molecular linear input specification corresponding to the drug molecule vector.

8. The method according to claim 1, characterized in that The method further comprises: The grammatical rules of the training samples are sampled by the mean vector and the covariance matrix in the encoder in the encoding model to obtain the feature vector of the training samples; Decoding the feature vector of the training sample by a decoder in the coding model to obtain a second grammar rule corresponding to the training sample; The model parameters of the encoding model are updated according to the grammatical rule of the training sample and the second grammatical rule of the training sample.

9. The method according to claim 8, characterized in that Before sampling the grammatical rules of the training samples by using the mean vector and the covariance matrix in the encoder in the encoding model to obtain the feature vector of the training samples, the method further includes: The grammatical rule sequence corresponding to the molecular linear input specification of the training sample is disassembled to obtain the grammatical rule of at least one training sample.

10. The method according to claim 8 or 9, characterized in that: The decoder is the nth decoder, The decoding of the feature vector of the training sample by the decoder in the coding model to obtain a second grammar rule corresponding to the training sample includes: Decoding the output of the n-1th encoder and the feature vector of the training sample by the nth decoder in the encoding model to obtain a second grammar rule corresponding to the training sample; Wherein, n is greater than or equal to 2 and less than or equal to N, and N is the number of decoders in the coding model.

11. The method according to claim 1, characterized in that: The method further comprises: Predicting the drug molecule vector of the training sample by a molecular property prediction model to obtain the drug property corresponding to the training sample; Determining the value of the loss function of the molecular property prediction model according to the drug properties of the training sample and the standard drug properties; According to the value of the loss function, the model parameters of the molecular property prediction model are updated.

12. A drug molecule processing device based on artificial intelligence, characterized in that: The device comprises: A first determination module, configured to determine a drug molecule property requirement in response to a generation request for a drug molecule; An acquisition module, used for acquiring randomly sampled drug molecule vectors, wherein the randomly sampled drug molecule vectors are obtained by performing random sampling processing on a set of drug molecule vectors obeying a multivariate Gaussian distribution; A prediction module, used for transforming the randomly sampled drug molecule vector to obtain a drug molecule vector that meets the drug molecule property requirements; An encoding module, used for decoding the drug molecule vector that meets the drug molecule property requirements to obtain a first grammatical rule corresponding to the drug molecule vector, wherein the first grammatical rule is used to represent a partial structure of the drug molecule in a vector or index manner; The second determination module is used to generate the drug molecule corresponding to the first grammatical rule according to the mapping relationship between the grammatical rule and the drug molecule, so as to respond to the generation request.

13. A drug molecule processing device based on artificial intelligence, characterized in that: The device comprises: A memory for storing executable instructions; A processor, configured to implement the artificial intelligence-based drug molecule processing method according to any one of claims 1 to 11 when executing executable instructions stored in the memory.

14. A storage medium, characterized in that: Executable instructions are stored, which are used to cause the processor to execute and implement the artificial intelligence-based drug molecule processing method described in any one of claims 1 to 11.

Images