Drug molecule generation and model training method, device and equipment

Abstract

The invention provides a drug molecule generation and model training method, device and equipment, and relates to the technical field of artificial intelligence, in particular to the technical fields of natural language processing, deep learning, computational biology, chemistry and the like. The drug molecule generation method comprises the following steps: generating candidate drug molecules and candidate synthesis routes corresponding to the candidate drug molecules based on a target spot and a pre-constructed real synthesis route; obtaining the affinity between the candidate drug molecule and the target spot; and determining a target drug molecule in the candidate drug molecules based on the candidate synthesis route and the affinity.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to the technical fields of natural language processing, deep learning, computational biology and chemistry, and in particular to a method, device and equipment for drug molecule generation and model training. Background Art

[0002] The molecular generation task is to iteratively generate unknown molecules on a given target to obtain small drug molecules with high target affinity. Summary of the invention

[0003] The present invention provides a method, device and apparatus for drug molecule generation and model training.

[0004] According to one aspect of the present disclosure, a method for generating a drug molecule is provided, comprising: generating candidate drug molecules and their corresponding candidate synthesis routes based on a target and a pre-constructed real synthesis route; obtaining the affinity between the candidate drug molecule and the target; and determining a target drug molecule among the candidate drug molecules based on the candidate synthesis route and the affinity.

[0005] According to another aspect of the present disclosure, a drug molecule generation model training method is provided, comprising: using the drug molecule generation model to process the input target target sample and the pre-constructed real synthesis route to output a predicted drug molecule and its corresponding predicted synthesis route; obtaining the predicted affinity between the predicted drug molecule and the target target sample; constructing a target loss function based on the predicted synthesis route and the predicted affinity; and adjusting the model parameters of the drug molecule generation model based on the target loss function.

[0006] According to another aspect of the present disclosure, a drug molecule generation device is provided, comprising: a generation module for generating candidate drug molecules and their corresponding candidate synthesis routes based on a target and a pre-constructed real synthesis route; an acquisition module for determining the affinity between the candidate drug molecule and the target; and a determination module for determining a target drug molecule among the candidate drug molecules based on the candidate synthesis route and the affinity.

[0007] According to another aspect of the present disclosure, a drug molecule generation model training device is provided, comprising: a generation module, used to use the drug molecule generation model to process the input target target sample and the pre-constructed real synthesis route to output the predicted drug molecule and its corresponding predicted synthesis route; an acquisition module, used to obtain the predicted affinity between the predicted drug molecule and the target target sample; a construction module, used to construct a target loss function based on the predicted synthesis route and the predicted affinity; and an adjustment module, used to adjust the model parameters of the drug molecule generation model based on the target loss function.

[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any method as described in any of the above aspects.

[0009] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any one of the methods according to any one of the above aspects.

[0010] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the methods described in any one of the above aspects.

[0011] According to the technical solution disclosed in the present invention, the synthesizability and accuracy of target drug molecules can be improved.

[0012] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0015] Figure 2 is a schematic diagram of an application scenario for implementing an embodiment of the present disclosure;

[0016] Figure 3 is a schematic diagram of a drug molecule generation model provided according to an embodiment of the present disclosure;

[0017] Figure 4is a schematic diagram according to a second embodiment of the present disclosure;

[0018] Figure 5 is a schematic diagram according to a third embodiment of the present disclosure;

[0019] Figure 6 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0020] Figure 7 is a schematic diagram according to a fifth embodiment of the present disclosure;

[0021] Figure 8 It is a schematic diagram of an electronic device used to implement the drug molecule generation method or drug molecule generation model training method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0023] In the related art, usually only the affinity between the generated drug molecules and the target is considered, which leads to the problem that the target drug molecules finally obtained may not be synthesized.

[0024] In order to improve the synthesizability and accuracy of target drug molecules, the present disclosure provides the following examples.

[0025] Figure 1 is a schematic diagram according to the first embodiment of the present disclosure. This embodiment provides a method for generating a drug molecule, the method comprising:

[0026] 101. Generate candidate drug molecules and their corresponding candidate synthetic routes based on the target and pre-constructed real synthetic routes.

[0027] 102. Obtain the affinity between the candidate drug molecule and the target.

[0028] 103. Based on the candidate synthesis route and the affinity, determine the target drug molecule among the candidate drug molecules.

[0029] Targets are proteins that drugs directly bind to, such as enzymes, ion channels and receptors or other biological molecules. Most drug targets are proteins, and the chemicals used to treat or diagnose diseases selectively interact with target proteins, causing changes in their biological pathways or functions.

[0030] The target site refers to the target site to be interacted with the target drug molecule and is a known quantity.

[0031] The synthesis route is used to indicate the synthesis process of drug molecules. Drug synthesis is the process of gradually converting raw material compounds into target drug molecules through chemical synthesis methods.

[0032] The real synthetic route refers to the synthetic route obtained based on wet experiments.

[0033] Among them, dry experiments refer to experiments conducted using computers and bioinformatics techniques, while wet experiments refer to experiments that process liquids and biological samples in the laboratory.

[0034] Since wet experiments are experiments conducted in the laboratory, they are verified by the laboratory and have practical synthesizability.

[0035] After obtaining the target and the actual synthesis route, they are input into the generative model, and the output includes candidate drug molecules and their corresponding candidate synthesis routes. The generative model is a pre-trained deep learning model, such as the Transformer model.

[0036] There is at least one real synthetic route, and based on each real synthetic route, at least one candidate drug molecule and its corresponding candidate synthetic route can be generated.

[0037] Generally speaking, each real synthetic route can generate multiple candidate drug molecules and their corresponding candidate synthetic routes. For example, based on the real synthetic route a, a first candidate drug molecule and its corresponding first candidate synthetic route, a second candidate drug molecule and its corresponding second candidate synthetic route, etc. can be generated.

[0038] For each candidate drug molecule, the affinity between the candidate drug molecule and the target can be calculated.

[0039] Affinity refers to the affinity between the drug and the target, reflecting the binding ability between the drug and the target. The greater the affinity, the tighter the binding between the two and the stronger the drug activity.

[0040] After obtaining the candidate synthesis route, the synthesizability of the corresponding candidate drug molecule can be analyzed based on the candidate synthesis route. For example, the candidate synthesis route is input into a pre-trained analysis model, and the output is the synthesizability result, which includes synthesizable or non-synthesizable.

[0041] For each candidate drug molecule, its corresponding affinity can be obtained, and based on the candidate synthesis route of the candidate drug molecule, it is analyzed whether the candidate drug molecule can be synthesized.

[0042] Afterwards, target drug molecules can be identified among the candidate drug molecules based on their synthesizability and their affinity to the target of interest.

[0043] Specifically, synthesizable candidate drug molecules can be obtained based on synthesizability, and then the candidate drug molecule with the greatest affinity can be selected as the target drug molecule.

[0044] In this embodiment, by obtaining the candidate synthesis route of the candidate drug molecule, the synthesizability of the candidate drug molecule can be analyzed based on the candidate synthesis route, thereby improving the synthesizability of the target drug molecule; in addition, by generating the candidate drug molecule based on the actual synthesis route, the accuracy of the candidate drug molecule can be improved, thereby improving the accuracy of the target drug molecule determined based on the candidate drug molecule.

[0045] In order to better understand the embodiments of the present disclosure, application scenarios to which the embodiments of the present disclosure can be applied are described.

[0046] Figure 2 It is a schematic diagram of an application scenario for implementing the embodiment of the present disclosure.

[0047] like Figure 2 As shown, an application (Application, APP) for drug molecule generation can be installed on the user terminal 201. The APP interacts with the server, which is deployed on the server 202. The server can be a local server or a cloud server, and can be a single server or a server cluster. The user terminal and the server can be connected via a wired network and / or a wireless network.

[0048] The user can send target information, such as target name, identifier or characterization sequence, to the server through the APP; the server obtains the characterization sequence of the target by querying or directly based on the information sent by the user terminal.

[0049] The server can also obtain pre-built real synthesis routes.

[0050] The actual synthetic route can be constructed based on single-step reaction data, which is known data, such as obtained from an open source single-step reaction database or a purchased commercial database.

[0051] After obtaining the single-step reaction data, the single-step reaction data can be spliced ​​according to the chemical reaction rules to obtain the real synthesis route. Generally speaking, the number of real synthesis routes is large, and a large number of candidate drug molecules can be generated.

[0052] After obtaining the target and the true synthetic route, the characterization sequence of the target and the true synthetic route are input into the drug molecule synthesis model, and the output is the candidate drug molecule and its corresponding candidate synthetic route.

[0053] After obtaining the candidate drug molecule, the affinity between the candidate drug molecule and the target is calculated. For example, a preset algorithm is used to calculate the docking score between the candidate drug molecule and the target. The larger the docking score, the stronger the affinity.

[0054] After obtaining the candidate synthesis route, the synthesizability of the candidate drug molecule is calculated. For example, the candidate synthesis route is input into a preset analysis model, and the output is a synthesizability result. The synthesizability result is used to indicate whether the candidate drug molecule corresponding to the candidate synthesis route is synthesizable or not.

[0055] After obtaining the synthesizability and affinity corresponding to the candidate drug molecules, the synthesizable candidate drug molecules with the greatest affinity can be used as target drug molecules.

[0056] The candidate drug molecules output by the drug molecule synthesis model can specifically be a molecular structure diagram, and the corresponding target drug molecules can also specifically be a molecular structure diagram, which refers to a graphical structure composed of atoms and chemical bonds. Afterwards, the server can feed back the molecular structure diagram of the target drug molecule to the APP on the user terminal, which will be displayed to the user by the APP. This embodiment takes the server-side processing as an example. It can be understood that if the user terminal itself has the corresponding capabilities, the relevant processes can also be executed locally on the user terminal.

[0057] Figure 3 It is a schematic diagram of a drug molecule generation model provided according to an embodiment of the present disclosure.

[0058] like Figure 3 As shown, the drug molecule generation model includes: a first embedding layer, a second embedding layer, an encoding layer and a decoding layer.

[0059] The first embedding layer is used to convert the target point into an embedding vector. Specifically, after the representation sequence of the target point is input into the first embedding layer, the output is the first embedding vector.

[0060] The second embedding layer is used to convert the real synthesis route into an embedding vector. Specifically, after the representation sequence of the real synthesis route is input into the second embedding layer, the output is the second embedding vector.

[0061] Generally speaking, there are multiple real synthetic routes. This embodiment takes the real synthetic route including: real synthetic route a and real synthetic route b as an example.

[0062] For each real synthesis route, the second embedding layer is used to convert it into a corresponding embedding vector, and different real synthesis routes can share the same second embedding layer.

[0063] After obtaining the first embedding vector and the second embedding vector, the first embedding vector and the second embedding vector corresponding to each real synthetic route are added to obtain a fusion vector corresponding to each real synthetic route. For example, corresponding to real synthetic route a, the first embedding vector and the second embedding vector corresponding to real synthetic route a are added to obtain a fusion vector corresponding to real synthetic route a.

[0064] After obtaining the fusion vector corresponding to each real synthetic route, the fusion vector of the encoding layer is used for encoding to obtain the hidden layer vector. Different real synthetic routes can share the same encoding layer. For example, corresponding to the real synthetic route a, the fusion vector corresponding to the real synthetic route a is encoded using the encoding layer to obtain the hidden layer vector corresponding to the real synthetic route a.

[0065] After obtaining the hidden vector corresponding to each real synthetic route, the decoding layer is used to decode the hidden vector to obtain the candidate drug molecules and their corresponding candidate synthetic routes.

[0066] For each real synthetic route, multiple candidate drug molecules and their corresponding candidate synthetic routes can be generated. For example, corresponding to the real synthetic route a, different candidate drug molecules and their corresponding candidate synthetic routes are obtained by using different decoding layers, such as using decoding layer-1 to decode the hidden layer vector corresponding to the real synthetic route a to obtain candidate drug molecule ad1 and its corresponding candidate synthetic route ar1, ..., using decoding layer-n to decode the hidden layer vector corresponding to the real synthetic route a to obtain candidate drug molecule adn and its corresponding candidate synthetic route arn.

[0067] Different real synthetic routes can share decoding layers or not. Taking sharing as an example, corresponding to the real synthetic route b, if the decoding layer-1 is used to decode the hidden layer vector corresponding to the real synthetic route b, the candidate drug molecule bd1 and its corresponding candidate synthetic route br1 are obtained,..., the decoding layer-n is used to decode the hidden layer vector corresponding to the real synthetic route b, and the candidate drug molecule bdn and its corresponding candidate synthetic route brn are obtained.

[0068] After obtaining multiple candidate drug molecules and their corresponding candidate synthesis routes, the candidate synthesis routes are used to determine the synthesizability of the candidate drug molecules, and the affinity of the candidate drug molecules to the target is calculated. After that, the synthesizable candidate drug molecules with the greatest affinity can be selected as the target drug molecules.

[0069] In combination with the above application scenarios, the present disclosure also provides the following embodiments.

[0070] Figure 4is a schematic diagram according to the second embodiment of the present disclosure. This embodiment provides a method for generating a drug molecule, the method comprising:

[0071] 401. Construct the actual synthesis route based on the pre-acquired single-step reaction data.

[0072] The single-step reaction data may be obtained from an open source single-step reaction database or a purchased commercial database.

[0073] After obtaining the single-step reaction data, the single-step reaction data can be spliced ​​according to the chemical reaction rules to obtain the real synthesis route. Generally speaking, the number of real synthesis routes is large, and a large number of candidate drug molecules can be generated.

[0074] In this embodiment, a true synthetic route is obtained based on single-step reaction data. Since the dimension of variation in the true synthetic route is much larger than the dimension of variation in the single-step reaction, a larger searchable molecular space can be generated based on the true synthetic route, thereby improving the accuracy of candidate drug molecules and thus improving the accuracy of target drug molecules. Single-step reaction data is experimentally verified to be feasible, and splicing rules are also feasible. Therefore, the true synthetic route obtained based on splicing rules and single-step reaction data has a high laboratory feasibility, thereby improving the synthesizability of target drug molecules.

[0075] 402. Using a pre-trained drug molecule generation model, the input target and the actual synthesis route are processed to output candidate drug molecules and candidate synthesis routes corresponding to the candidate drug molecules.

[0076] Among them, the drug molecule generation model is a deep learning model. Using this model to generate drug molecules can take advantage of the excellent performance of the deep learning model to improve the accuracy and generation efficiency of candidate drug molecules and their candidate synthesis routes.

[0077] The training process of the drug molecule generation module can refer to the description of the subsequent related embodiments.

[0078] In some embodiments, the drug molecule generation model includes: a first embedding layer, a second embedding layer, an encoding layer, and a decoding layer; the pre-trained drug molecule generation model is used to process the input target and the real synthesis route to output the candidate drug molecule and its corresponding candidate synthesis route, including:

[0079] The first embedding layer is used to embed the target to obtain a first embedding vector; the second embedding layer is used to embed the real synthesis route to obtain a second embedding vector; the first embedding vector and the second embedding vector are added to obtain a fusion vector; the encoding layer is used to encode the fusion vector to obtain a hidden layer vector; the decoding layer is used to decode the hidden layer vector to obtain the candidate drug molecule and the candidate synthesis route.

[0080] In this embodiment, based on the processing of the above layers, candidate drug molecules and their corresponding candidate synthesis routes can be obtained simply and efficiently.

[0081] 403. Obtain the affinity between the candidate drug molecule and the target.

[0082] For example, a candidate drug molecule and a target are input into a docking score calculation model, and the output is a docking score, based on which affinity is determined, i.e., a higher docking score indicates a greater affinity.

[0083] 404. Determine synthesizable candidate drug molecules based on the candidate synthesis route; and, among the synthesizable candidate drug molecules, use the candidate drug molecule with the greatest affinity as the target drug molecule.

[0084] After obtaining the candidate synthesis route, the candidate synthesis route can be input into a preset analysis model. The analysis model can specifically be a binary classification model, which indicates whether it is synthesizable or not synthesizable.

[0085] For a candidate synthesis route, if the analysis result is that it is synthesizable, it indicates that the candidate drug molecule corresponding to the candidate synthesis route is synthesizable. Conversely, if the analysis result is that it is not synthesizable, it indicates that the candidate drug molecule corresponding to the candidate synthesis route is not synthesizable.

[0086] After obtaining the synthesizable candidate drug molecules based on the candidate synthesis route, the candidate drug molecule with the greatest affinity to the target can be selected as the target drug molecule. For example, if the first candidate drug molecule and the second candidate drug are both synthesizable, and the affinity of the first candidate drug molecule to the target target is greater than the affinity of the second candidate drug molecule to the target target, the first candidate drug molecule is selected as the target drug molecule.

[0087] In this embodiment, among the synthesizable candidate drug molecules, the candidate drug molecule with the greatest affinity is used as the target drug molecule, so that a target drug molecule with greater affinity and synthesizable can be obtained, thereby improving the performance of the target drug molecule.

[0088] Figure 5is a schematic diagram according to the third embodiment of the present disclosure. This embodiment provides a drug molecule generation model training method, the training method comprising:

[0089] 501. A drug molecule generation model is used to process the input target samples and pre-constructed real synthesis routes to output predicted drug molecules and their corresponding predicted synthesis routes.

[0090] 502. Obtain predicted affinity between the predicted drug molecule and the target sample.

[0091] 503. Construct a target loss function based on the predicted synthesis route and the predicted affinity.

[0092] 504. Based on the target loss function, adjust the model parameters of the drug molecule generation model.

[0093] The target sample is predetermined, and the real synthetic route is pre-constructed, for example, the real synthetic route is constructed based on pre-acquired single-step reaction data.

[0094] Using a similar process to the above reasoning process, during the training phase, the output of the drug molecule generation model is called predicted drug molecules and predicted synthesis routes, and the affinity between the predicted drug molecules and the target samples is called predicted affinity.

[0095] After obtaining the predicted synthesis route and predicted affinity, the objective function can be constructed, and then the model parameters can be adjusted based on the target loss function, such as adjusting the model's weight coefficient w and bias coefficient b, until the preset end condition is reached. The model that reaches the preset end condition is used as the final model.

[0096] The drug molecule generation model finally generated can be used in the reasoning stage to obtain candidate drug molecules and their corresponding candidate synthesis routes.

[0097] In this embodiment, by obtaining the predicted synthesis route of the predicted drug molecule, constructing a target loss function based on the predicted synthesis route and the predicted affinity, and adjusting the model parameters based on the target loss function, a model with high synthesizability and accuracy can be obtained, and then when the model is used for drug generation, the synthesizability and accuracy of the generated drug molecules can be improved.

[0098] In some embodiments, the predicted drug molecules can be divided into positive samples and negative samples based on the predicted synthesis route; based on the predicted affinity corresponding to the positive samples and the predicted affinity corresponding to the negative samples, a comparative loss function is constructed as the target loss function.

[0099] For example, after obtaining the predicted synthesis route, it can be input into the analysis model, and the output is the synthesizability result, and then the predicted drug molecules are divided into positive samples and negative samples based on the synthesizability result. For example, the predicted drug molecules that cannot be synthesized are used as negative samples, and the predicted drug molecules that can be synthesized are used as positive samples.

[0100] Afterwards, we can use contrastive learning to construct a contrastive loss function using the predicted affinity of positive and negative samples and use it as the target loss function.

[0101] In this embodiment, positive and negative samples are distinguished based on the predicted synthesis route, and then a target loss function is constructed based on the predicted affinity of the positive and negative samples. The target loss function can be obtained by contrastive learning to improve the model performance.

[0102] In some embodiments, the drug molecule generation model includes: a first embedding layer, a second embedding layer, an encoding layer, and a decoding layer; the drug molecule generation model is used to process the input target sample and the pre-constructed real synthesis route to output the predicted drug molecule and its corresponding predicted synthesis route, including:

[0103] The first embedding layer is used to embed the target sample to obtain a first embedding vector; the second embedding layer is used to embed the real synthesis route to obtain a second embedding vector; the first embedding vector and the second embedding vector are added to obtain a fusion vector; the encoding layer is used to encode the fusion vector to obtain a hidden layer vector; the decoding layer is used to decode the hidden layer vector to obtain the predicted drug molecule and the predicted synthesis route.

[0104] In this embodiment, based on the processing of the above layers, the predicted drug molecules and their corresponding predicted synthesis routes can be obtained simply and efficiently.

[0105] Figure 6 6 is a schematic diagram according to the fourth embodiment of the present disclosure. This embodiment provides a drug molecule generation device, the device 600 includes: a generation module 601, an acquisition module 602 and a determination module 603.

[0106] The generation module 601 is used to generate candidate drug molecules and their corresponding candidate synthesis routes based on the target and the pre-constructed real synthesis route; the acquisition module 602 is used to determine the affinity between the candidate drug molecules and the target; the determination module 603 is used to determine the target drug molecule among the candidate drug molecules based on the candidate synthesis route and the affinity.

[0107] In this embodiment, by obtaining the candidate synthesis route of the candidate drug molecule, the synthesizability of the candidate drug molecule can be analyzed based on the candidate synthesis route, thereby improving the synthesizability of the target drug molecule; in addition, by generating the candidate drug molecule based on the actual synthesis route, the accuracy of the candidate drug molecule can be improved, thereby improving the accuracy of the target drug molecule determined based on the candidate drug molecule.

[0108] In some embodiments, the determining module 603 is further configured to:

[0109] Determining candidate drug molecules that can be synthesized based on the candidate synthetic route;

[0110] Among the synthesizable candidate drug molecules, the candidate drug molecule with the greatest affinity is used as the target drug molecule.

[0111] In this embodiment, among the synthesizable candidate drug molecules, the candidate drug molecule with the greatest affinity is used as the target drug molecule, so that a target drug molecule with greater affinity and synthesizable can be obtained, thereby improving the performance of the target drug molecule.

[0112] In some embodiments, the apparatus 600 further includes:

[0113] A construction module is used to construct the real synthesis route based on the pre-acquired single-step reaction data.

[0114] In this embodiment, a true synthetic route is obtained based on single-step reaction data. Since the dimension of variation in the true synthetic route is much larger than the dimension of variation in the single-step reaction, a larger searchable molecular space can be generated based on the true synthetic route, thereby improving the accuracy of candidate drug molecules and thus improving the accuracy of target drug molecules. Single-step reaction data is experimentally verified to be feasible, and splicing rules are also feasible. Therefore, the true synthetic route obtained based on splicing rules and single-step reaction data has a high laboratory feasibility, thereby improving the synthesizability of target drug molecules.

[0115] In some embodiments, the generating module 601 is further used to:

[0116] The pre-trained drug molecule generation model is used to process the input target and the real synthesis route to output the candidate drug molecule and the candidate synthesis route.

[0117] Among them, the drug molecule generation model is a deep learning model. Using this model to generate drug molecules can take advantage of the excellent performance of the deep learning model to improve the accuracy and generation efficiency of candidate drug molecules and their candidate synthesis routes.

[0118] In some embodiments, the drug molecule generation model includes: a first embedding layer, a second embedding layer, an encoding layer, and a decoding layer; the generation module 601 is further used to:

[0119] The first embedding layer is used to embed the input target point to obtain a first embedding vector; the second embedding layer is used to embed the real synthesis route to obtain a second embedding vector; the first embedding vector and the second embedding vector are added to obtain a fusion vector; the encoding layer is used to encode the fusion vector to obtain a hidden layer vector; the decoding layer is used to decode the hidden layer vector to output the candidate drug molecule and the candidate synthesis route.

[0120] In this embodiment, based on the processing of the above layers, candidate drug molecules and their corresponding candidate synthesis routes can be obtained simply and efficiently.

[0121] Figure 7 Schematic diagram of the fifth embodiment of the present disclosure. This embodiment provides a drug molecule generation model training device, the device 700 includes: a generation module 701, an acquisition module 702, a construction module 703 and an adjustment module 704.

[0122] The generation module 701 is used to use the drug molecule generation model to process the input target target sample and the pre-constructed real synthesis route to output the predicted drug molecule and its corresponding predicted synthesis route; the acquisition module 702 is used to obtain the predicted affinity between the predicted drug molecule and the target target sample; the construction module 703 is used to construct a target loss function based on the predicted synthesis route and the predicted affinity; the adjustment module 704 is used to adjust the model parameters of the drug molecule generation model based on the target loss function.

[0123] In this embodiment, by obtaining the predicted synthesis route of the predicted drug molecule, constructing a target loss function based on the predicted synthesis route and the predicted affinity, and adjusting the model parameters based on the target loss function, a model with high synthesizability and accuracy can be obtained, and then when the model is used for drug generation, the synthesizability and accuracy of the generated drug molecules can be improved.

[0124] In some embodiments, the construction module 703 is further used to:

[0125] Based on the predicted synthesis route, the predicted drug molecules are divided into positive samples and negative samples; based on the predicted affinity corresponding to the positive samples and the predicted affinity corresponding to the negative samples, a comparative loss function is constructed as the target loss function.

[0126] In this embodiment, positive and negative samples are distinguished based on the predicted synthesis route, and then a target loss function is constructed based on the predicted affinity of the positive and negative samples. The target loss function can be obtained by contrastive learning to improve the model performance.

[0127] In some embodiments, the drug molecule generation model includes: a first embedding layer, a second embedding layer, an encoding layer, and a decoding layer; the generation module 701 is further used to:

[0128] The first embedding layer is used to embed the input target sample to obtain a first embedding vector; the second embedding layer is used to embed the real synthesis route to obtain a second embedding vector; the first embedding vector and the second embedding vector are added to obtain a fusion vector; the encoding layer is used to encode the fusion vector to obtain a hidden layer vector; the decoding layer is used to decode the hidden layer vector to output the predicted drug molecule and the predicted synthesis route.

[0129] In this embodiment, based on the processing of the above layers, the predicted drug molecules and their corresponding predicted synthesis routes can be obtained simply and efficiently.

[0130] It can be understood that in the embodiments of the present disclosure, the same or similar contents in different embodiments can be referenced to each other.

[0131] It can be understood that the “first”, “second”, etc. in the embodiments of the present disclosure are only used for distinction and do not indicate the degree of importance, time sequence, etc.

[0132] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0133] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0134] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device 800 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 800 can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0135] like Figure 8 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the electronic device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0136] Multiple components in the electronic device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the electronic device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0137] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above, such as a drug molecule generation method or a drug molecule generation model training method. For example, in some embodiments, the drug molecule generation method or the drug molecule generation model training method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 800 via ROM 802 and / or a communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the drug molecule generation method or the drug molecule generation model training method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to execute the drug molecule generation method or the drug molecule generation model training method in any other appropriate manner (eg, by means of firmware).

[0138] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0139] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0140] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0141] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0142] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0143] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0144] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0145] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for generating a drug molecule, comprising: Generate candidate drug molecules and their corresponding candidate synthetic routes based on the target and pre-built real synthetic routes; Obtaining the affinity between the candidate drug molecule and the target; Based on the candidate synthesis routes and the affinity, a target drug molecule is determined among the candidate drug molecules.

2. The method according to claim 1, wherein: The step of determining a target drug molecule among the candidate drug molecules based on the candidate synthesis route and the affinity comprises: Determining candidate drug molecules that can be synthesized based on the candidate synthetic route; Among the synthesizable candidate drug molecules, the candidate drug molecule with the greatest affinity is used as the target drug molecule.

3. The method according to claim 1, further comprising: The actual synthetic route is constructed based on the single-step reaction data acquired in advance.

4. The method according to claim 1, wherein: The method of generating candidate drug molecules and their corresponding synthetic routes based on the target and the pre-constructed real synthetic routes includes: The pre-trained drug molecule generation model is used to process the input target and the real synthesis route to output the candidate drug molecule and the candidate synthesis route.

5. The method according to claim 4, wherein: The drug molecule generation model comprises: a first embedding layer, a second embedding layer, an encoding layer and a decoding layer; The method uses a pre-trained drug molecule generation model to process the input target and the real synthesis route to output the candidate drug molecule and its corresponding candidate synthesis route, including: Using the first embedding layer, embedding the input target point to obtain a first embedding vector; Embedding the real synthesis route using the second embedding layer to obtain a second embedding vector; Adding the first embedding vector and the second embedding vector to obtain a fused vector; Using the encoding layer, encoding the fusion vector to obtain a hidden layer vector; The decoding layer is used to decode the hidden layer vector to output the candidate drug molecules and the candidate synthesis routes.

6. A drug molecule generation model training method, comprising: The drug molecule generation model is used to process the input target samples and the pre-built real synthesis routes to output the predicted drug molecules and their corresponding predicted synthesis routes; Obtaining the predicted affinity between the predicted drug molecule and the target sample; constructing a target loss function based on the predicted synthetic route and the predicted affinity; Based on the objective loss function, the model parameters of the drug molecule generation model are adjusted.

7. The method according to claim 6, wherein: The step of constructing a loss function based on the predicted synthesis route and the predicted affinity comprises: Based on the predicted synthesis route, the predicted drug molecules are divided into positive samples and negative samples; Based on the predicted affinity corresponding to the positive sample and the predicted affinity corresponding to the negative sample, a contrast loss function is constructed as the target loss function.

8. The method according to claim 6, wherein: The drug molecule generation model comprises: a first embedding layer, a second embedding layer, an encoding layer and a decoding layer; The drug molecule generation model is used to process the input target sample and the pre-constructed real synthesis route to output the predicted drug molecule and its corresponding predicted synthesis route, including: Using the first embedding layer, embedding the input target point sample to obtain a first embedding vector; Embedding the real synthesis route using the second embedding layer to obtain a second embedding vector; Adding the first embedding vector and the second embedding vector to obtain a fused vector; Using the encoding layer, encoding the fusion vector to obtain a hidden layer vector; The decoding layer is used to decode the hidden layer vector to output the predicted drug molecule and the predicted synthesis route.

9. A drug molecule generation device, comprising: A generation module is used to generate candidate drug molecules and their corresponding candidate synthesis routes based on the target and the pre-built real synthesis route; An acquisition module, used to determine the affinity between the candidate drug molecule and the target; A determination module is used to determine a target drug molecule among the candidate drug molecules based on the candidate synthesis route and the affinity.

10. The device according to claim 9, wherein: The determination module is further used for: Determining candidate drug molecules that can be synthesized based on the candidate synthetic route; Among the synthesizable candidate drug molecules, the candidate drug molecule with the greatest affinity is used as the target drug molecule.

11. The apparatus according to claim 9, further comprising: A construction module is used to construct the real synthesis route based on the pre-acquired single-step reaction data.

12. The device according to claim 9, wherein: The generating module is further used for: The pre-trained drug molecule generation model is used to process the input target and the real synthesis route to output the candidate drug molecule and the candidate synthesis route.

13. The device according to claim 12, wherein: The drug molecule generation model comprises: a first embedding layer, a second embedding layer, an encoding layer and a decoding layer; The generating module is further used for: Embedding the input target point using the first embedding layer to obtain a first embedding vector; Using the second embedding layer, embedding the real synthesis route to obtain a second embedding vector; Adding the first embedding vector and the second embedding vector to obtain a fused vector; Using the encoding layer, encoding the fusion vector to obtain a hidden layer vector; The decoding layer is used to decode the hidden layer vector to output the candidate drug molecules and the candidate synthesis routes.

14. A drug molecule generation model training device, comprising: A generation module, for processing the input target sample and the pre-built real synthesis route using the drug molecule generation model, so as to output the predicted drug molecule and its corresponding predicted synthesis route; An acquisition module, used to acquire the predicted affinity between the predicted drug molecule and the target sample; A construction module, for constructing a target loss function based on the predicted synthetic route and the predicted affinity; An adjustment module is used to adjust the model parameters of the drug molecule generation model based on the target loss function.

15. The device according to claim 14, wherein: The building blocks are further used to: Based on the predicted synthesis route, the predicted drug molecules are divided into positive samples and negative samples; Based on the predicted affinity corresponding to the positive sample and the predicted affinity corresponding to the negative sample, a contrast loss function is constructed as the target loss function.

16. The device according to claim 14, wherein: The drug molecule generation model comprises: a first embedding layer, a second embedding layer, an encoding layer and a decoding layer; The generating module is further used for: Using the first embedding layer, embedding the input target point sample to obtain a first embedding vector; Using the second embedding layer, embedding the real synthesis route to obtain a second embedding vector; Adding the first embedding vector and the second embedding vector to obtain a fused vector; Using the encoding layer, encoding the fusion vector to obtain a hidden layer vector; The decoding layer is used to decode the hidden layer vector to output the predicted drug molecule and the predicted synthesis route.

17. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

18. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-8.

19. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images