Compound property prediction model training method, apparatus, device, and storage medium

Abstract

The present disclosure provides a compound property prediction model training method, device, equipment, storage medium and program product, relating to the technical field of artificial intelligence, in particular to the technical field of deep learning. A specific embodiment of the method comprises: obtaining an unlabeled compound dataset; pre-training a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network; obtaining a plurality of labeled compound datasets, wherein one labeled compound dataset labels one property of a compound; and performing multi-task training on the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model, wherein the compound property prediction model is used to predict multiple properties of a compound. The embodiment proposes a multi-stage compound pre-training and knowledge transfer training framework, which significantly improves the model effect.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, in particular to the technical field of deep learning. BACKGROUND

[0002] Based on long-term drug research and development practice, researchers found that the performance of candidate compounds in later clinical experiments is largely related to their Absorption, Distribution, Metabolism, Excretion and Toxicity in the human body. Therefore, accurately predicting the ADMET properties of candidate compounds can help quickly screen potential compounds with higher success rates at an early stage of drug research and development, thereby saving new drug development cycles and reducing research costs.

[0003] The core of the ADMET prediction platform is the algorithm model. Existing ADMET prediction platforms usually apply a certain deep learning model. However, deep learning models need to obtain a large amount of labeled training data for supervised training, and insufficient training data will limit the improvement of the model effect. SUMMARY

[0004] Embodiments of the present disclosure provide a compound property prediction model training method, device, equipment, storage medium and program product.

[0005] In a first aspect, embodiments of the present disclosure provide a compound property prediction model training method, comprising: obtaining an unlabeled compound dataset; pre-training a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network; obtaining a plurality of labeled compound datasets, wherein one labeled compound dataset labels one property of a compound; and performing multi-task training on the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model, wherein the compound property prediction model is used to predict multiple properties of a compound.

[0006] In a second aspect, embodiments of the present disclosure provide a compound property prediction model training device, comprising: a first obtaining module configured to obtain an unlabeled compound dataset; a first training module configured to pre-train a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network; a second obtaining module configured to obtain a plurality of labeled compound datasets, wherein one labeled compound dataset labels one property of a compound; and a second training module configured to perform multi-task training on the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model, wherein the compound property prediction model is used to predict multiple properties of a compound.

[0007] In a third aspect, the embodiments of the present disclosure provide an electronic device, comprising: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable 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 described in any one of the implementations of the first aspect.

[0008] In a fourth aspect, the embodiments of the present disclosure provide a non-transitory computer-readable storage medium storing computer instructions, the computer instructions being used to cause a computer to perform the method described in any one of the implementations of the first aspect.

[0009] In a fifth aspect, the embodiments of the present disclosure provide a computer program product comprising a computer program, the computer program being executed by a processor to implement the method described in any one of the implementations of the first aspect.

[0010] The compound property prediction model training method provided by the embodiments of the present disclosure proposes a multi-stage compound pre-training and knowledge transfer training framework. On the one hand, through a self-supervised training task, a model is pre-trained on a large-scale unlabeled compound dataset, so that the model learns the physical and chemical knowledge implied in the compound structure; on the other hand, through multi-task training, the model learns more extensive domain knowledge and realizes knowledge transfer. Thus, the model effect is significantly improved.

[0011] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0012] Other features, objects, and advantages of the present disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the drawings. The drawings are intended to better understand the present scheme and do not constitute a limitation of the present disclosure. Among them:

[0013] Figure 1 is a flowchart of some embodiments of the compound property prediction model training method according to the present disclosure;

[0014] Figure 2 is a flowchart of yet some embodiments of the compound property prediction model training method according to the present disclosure;

[0015] Figure 3 is a flowchart of still some embodiments of the compound property prediction model training method according to the present disclosure;

[0016] Figure 4 is a three-stage compound pre-training and property prediction framework;

[0017] Figure 5 is a structural diagram of some embodiments of a compound property prediction model training device according to the present disclosure;

[0018] Figure 6 is a block diagram of an electronic device for implementing a compound property prediction model training method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding, which should be considered in their context only. Thus, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0020] It should be noted that the embodiments in the present disclosure and the features in the embodiments can be combined with each other without conflict. The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0021] Figure 1 Flow 100 of some embodiments of a compound property prediction model training method according to the present disclosure is shown. The compound property prediction model training method includes the following steps:

[0022] Step 101, obtaining an unlabeled compound dataset.

[0023] In this embodiment, the execution subject of the compound property prediction model training method can obtain an unlabeled compound dataset.

[0024] Among them, the unlabeled compound dataset can include large-scale unlabeled labeled compound data. Compound data can include, but are not limited to, spatial structure, topology, sequence expression, chemical knowledge, etc. of the compound.

[0025] Step 102, pre-training a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network.

[0026] In this embodiment, the above execution subject can pre-train a graph neural network (GNN, Graph Neural Network) using the unlabeled compound dataset to obtain a pre-trained graph neural network.

[0027] Generally, pre-training techniques popular in the field of natural language processing and image processing can be used to design a variety of self-supervised tasks that can be applied to compound graph structures, and the graph neural network can be pre-trained on a large-scale unlabeled compound dataset, so that the model can learn the physical and chemical knowledge contained in the compound structure itself.

[0028] In some embodiments, the above execution subject can first obtain self-supervised information based on the unlabeled compound data in the unlabeled compound dataset; then pre-train the graph neural network by taking the unlabeled compound data as input and the self-supervised information as output, to obtain the pre-trained graph neural network. The self-supervised information can be the self-supervised information mined from the compound data. Through the self-supervised information, various self-supervised tasks applied to the compound graph structure can be designed, and the graph neural network can be pre-trained on a large-scale unlabeled compound dataset, so that the graph neural network can learn the physical and chemical knowledge contained in the compound structure itself. Here, the self-supervised information can include but is not limited to: compound local structure (Context), compound bond length (Bond length), compound bond angle (Bond angle), molecular fingerprints (Fingerprints), etc., which are mined from the sequence expression in the compound data by using existing chemical tools. Correspondingly, the self-supervised tasks designed based on the self-supervised information can include but are not limited to: compound local structure prediction, compound bond length prediction, compound bond angle prediction, molecular fingerprint prediction, etc.

[0029] Step 103, obtaining a plurality of labeled compound datasets.

[0030] In this embodiment, the above execution subject can obtain a plurality of labeled compound datasets.

[0031] Among the labeled compound datasets, there can be compound data labeled with property labels. One labeled compound dataset labels one property of the compound. Here, the plurality of labeled compound datasets can include but are not limited to at least two of the following: a compound dataset labeled with ADMET properties, a compound dataset labeled with biological activity, a compound dataset labeled with physicochemical properties of compounds, etc.

[0032] It should be noted that the size of the labeled compound dataset is much smaller than that of the unlabeled compound dataset, and the size of the compound dataset labeled with ADMET properties is much smaller than that of the compound dataset labeled with biological activity and the compound dataset labeled with physicochemical properties.

[0033] Step 104, multi-task training the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model.

[0034] In this embodiment, the above execution subject can multi-task train the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model. The compound property prediction model can be used to predict multiple properties of the compound.

[0035] Generally, multiple annotated compound datasets are combined for multitask training, so that the model learns more extensive domain knowledge and achieves knowledge transfer.

[0036] In some embodiments, in addition to the compound dataset annotated for ADMET properties, other annotated compound datasets of a larger scale can also be collected, such as a compound dataset annotated for other biological activities, a compound dataset annotated for physical and chemical properties of a compound, and the like. Combining these types of annotated compound datasets for multitask training enables the model to learn correlations from multiple compound datasets of physical, chemical, biological activity, biological toxicity, and the like, thereby achieving knowledge transfer. In particular, introducing a compound dataset annotated for other biological activities can significantly improve the model effect.

[0037] It should be noted that the embodiments of the present application can provide a more flexible multitask training process. The specific number of rounds of each task training is set to achieve flexible introduction and exit of tasks, thereby breaking the boundary between pre-training and downstream task fine-tuning. The first introduced task can be considered as pre-training in the traditional sense, and the later introduced task can be considered as downstream fine-tuning in the traditional sense. In this way, on the one hand, various custom tasks can be easily compatible, and other physical, chemical, and biological related datasets can be introduced to learn more extensive domain knowledge; on the other hand, by arranging the order of different tasks, the model can gradually focus on the downstream task that is really concerned.

[0038] The compound property prediction model training method provided by the embodiments of the present disclosure proposes a multi-stage compound pre-training and knowledge transfer training framework. On the one hand, through a self-supervised training task, the model is pre-trained on a large-scale unlabeled compound dataset, so that the model learns the physical and chemical knowledge implied in the compound structure; on the other hand, through multitask training, the model learns more extensive domain knowledge and achieves knowledge transfer. Thus, the model effect is significantly improved. Moreover, compared with existing supervised training, the cost of obtaining training data is reduced. The compound property prediction model trained to predict ADMET properties can be applied to virtual screening in the early stage of drug research and development, and can more accurately screen drugable compounds from a candidate compound library, thereby greatly reducing the risk of drug research and development failure, reducing costs, and thus greatly accelerating new drug discovery and development. The large-scale compound pre-training technology is applied to the compound ADMET property prediction task, and based on the multi-stage and multitask training framework, the prediction accuracy of ADMET properties is greatly improved.

[0039] With reference to Figure 2FIG. 20 shows a flow 200 illustrating yet some embodiments of the compound property prediction model training method according to the present disclosure. The compound property prediction model training method comprises the following steps:

[0040] Step 201, obtaining an unlabeled compound dataset.

[0041] Step 202, pre-training a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network.

[0042] Step 203, obtaining a plurality of labeled compound datasets.

[0043] Step 204, multi-task training the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model.

[0044] In the present embodiment, the specific operations of steps 201-204 have been described in detail in steps 101-104 of the embodiment shown in Figure 1

[0045] Step 205, obtaining a target labeled compound dataset.

[0046] In the present embodiment, the execution subject of the compound property prediction model training method can obtain a target labeled compound dataset.

[0047] The target labeled compound dataset can label the target property of the compound. In actual applications, the target labeled compound dataset can be a private dataset uploaded by a user. Generally, by providing a model fine-tuning interface that can be called by the user, the user can upload a private dataset.

[0048] Step 206, fine-tuning the compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model.

[0049] In the present embodiment, the execution subject described above can fine-tune the compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model. The fine-tuned compound property prediction model can be used to predict the target property of the compound.

[0050] Here, based on the multi-stage compound pre-training and knowledge transfer training framework, a large-scale pre-training model fine-tuning based on ADMET prediction tasks is added. Through the model fine-tuning interface, a private dataset is uploaded to fine-tune the compound property prediction model.

[0051] As can be seen from Figure 2 , compared with Figure 1 ​Compared with the corresponding embodiments, the process 200 of the compound property prediction model training method in this embodiment adds a model fine-tuning step. Therefore, the scheme described in this embodiment provides a model fine-tuning function on the ADMET prediction task based on the multi-stage compound pre-training and knowledge transfer training framework. By uploading a private dataset through the model fine-tuning interface, the compound property prediction model is fine-tuned, and the model effect is obviously improved compared with the model trained from random initialization.

[0052] Further reference Figure 3 which shows a process 300 of another embodiment of the compound property prediction model training method according to the present disclosure. The compound property prediction model training method comprises the following steps:

[0053] Step 301, obtaining an unlabeled compound dataset.

[0054] Step 302, pre-training a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network.

[0055] Step 303, obtaining a plurality of labeled compound datasets.

[0056] Step 304, multi-task training the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model.

[0057] In this embodiment, the specific operations of steps 301-304 have been described in detail in the embodiment shown in Figure 1 steps 101-104, which will not be repeated here.

[0058] Step 305, optimizing the compound property prediction model using one of the plurality of labeled compound datasets to obtain an optimized compound property prediction model.

[0059] In this embodiment, the above execution subject can optimize the compound property prediction model using one of the plurality of labeled compound datasets to obtain an optimized compound property prediction model. The optimized compound property prediction model can be used to predict one property of a compound.

[0060] Generally, after pre-training and multi-task training, the shared model parameters of the compound property prediction model can be copied and re-considered as an independent model, and continue to be trained in a single task mode for several rounds, so that the model parameters are further optimized for downstream tasks, thereby improving the model effect. For example, the compound property prediction model is further trained for several rounds using a compound dataset labeled with ADMET properties, so that the optimized compound property prediction model is used to predict the properties of ADMET of the compound. For another example, the compound property prediction model is further trained for several rounds using a compound dataset labeled with other biological activities, so that the optimized compound property prediction model is used to predict other biological activities of the compound.

[0061] Step 306, obtaining a target labeled compound dataset.

[0062] In this embodiment, the execution subject of the compound property prediction model training method can obtain a target labeled compound dataset.

[0063] The target labeled compound dataset can label the target property of the compound. In actual application, the target labeled compound dataset can be a private dataset uploaded by the user. Generally, by providing a model fine-tuning interface that can be called by the user, the user can upload a private dataset.

[0064] Step 307, fine-tuning the optimized compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model.

[0065] In this embodiment, the execution subject described above can fine-tune the compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model. The fine-tuned compound property prediction model can be used to predict the target property of the compound.

[0066] Here, based on the multi-stage compound pre-training and knowledge transfer training framework, fine-tuning based on large-scale pre-training models is added on the ADMET prediction task. Through the model fine-tuning interface, a private dataset is uploaded to fine-tune the compound property prediction model.

[0067] From Figure 3 It can be seen that, compared with Figure 1Compared with the corresponding embodiment, the process 300 of the compound property prediction model training method in the embodiment adds a model optimization step and a model fine-tuning step. Thus, the scheme described in the embodiment continues to train several rounds in a single task mode after pre-training and multi-task training, so that the model parameters are further optimized for downstream tasks, thereby improving the model effect. Moreover, based on the multi-stage compound pre-training and knowledge transfer training framework, the model fine-tuning function is provided on the ADMET prediction task. By uploading a private dataset through the model fine-tuning interface, the compound property prediction model is fine-tuned, and compared with the model trained from random initialization, the model effect is obviously improved.

[0068] For ease of understanding, Figure 4 A three-stage compound pre-training and property prediction framework is shown. As shown in Figure 4 The first stage is multi-task pre-training of unlabeled data. In the first stage, five self-supervised tasks that can be applied to the compound graph structure are designed, including compound local structure prediction, compound molecular fingerprint prediction, compound bond length prediction, compound bond angle prediction, etc. The GNN backbone network is pre-trained on a large-scale unlabeled compound dataset, so that the model learns the physical and chemical knowledge contained in the compound structure itself. The second stage is multi-task fine-tuning on labeled data. In the second stage, compound datasets labeled with ADMET properties, compound datasets labeled with other biological activities, compound datasets labeled with physical and chemical properties of compounds, etc. are collected to construct 40 fine-tuning tasks for multi-task training, so that the model learns the correlation from multiple datasets such as physical, chemical, biological activity, and biological toxicity, thereby realizing knowledge transfer. The third stage is single-task fine-tuning on ADMET endpoints. In the third stage, the shared model parameters are copied and re-considered as independent models, and 36 training nodes are constructed for properties such as Caco-2 permeability, BBBP, CYP450 1A2 inhibitor, and pKa. Continue to train several rounds in a single task mode, so that the model parameters are further optimized for downstream tasks, thereby improving the effect.

[0069] Further referring to Figure 5 As an implementation of the method shown in the above figures, the present disclosure provides some embodiments of a compound property prediction model training device, which device embodiments correspond to the method embodiments shown in Figure 1 The device can be applied in various electronic devices.

[0070] As Figure 5As shown, the compound property prediction model training apparatus 500 of the present embodiment can include a first obtaining module 501, a first training module 502, a second obtaining module 503, and a second training module 504. The first obtaining module 501 is configured to obtain an unlabeled compound dataset. The first training module 502 is configured to pre-train a graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network. The second obtaining module 503 is configured to obtain a plurality of labeled compound datasets, wherein one labeled compound dataset labels one property of a compound. The second training module 504 is configured to perform multi-task training on the pre-trained graph neural network using the plurality of labeled compound datasets to obtain a compound property prediction model, wherein the compound property prediction model is used to predict a plurality of properties of a compound.

[0071] In the present embodiment, the specific processes of the first obtaining module 501, the first training module 502, the second obtaining module 503, and the second training module 504 in the compound property prediction model training apparatus 500 and the technical effects brought by the specific processes can be respectively referred to the specific processes of the first obtaining module 501, the first training module 502, the second obtaining module 503, and the second training module 504 and the technical effects brought by the specific processes in the corresponding embodiments. Figure 1 The related descriptions of steps 101-104 in the corresponding embodiments will not be repeated here.

[0072] In some optional implementation manners of the present embodiment, the compound property prediction model training apparatus 500 further includes a third obtaining module configured to obtain a target labeled compound dataset, wherein the target labeled compound dataset labels a target property of a compound; and a first fine-tuning module configured to fine-tune the compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model, wherein the fine-tuned compound property prediction model is used to predict the target property of the compound.

[0073] In some optional implementation manners of the present embodiment, the compound property prediction model training apparatus 500 further includes an optimization module configured to optimize the compound property prediction model using one labeled compound dataset in the plurality of labeled compound datasets to obtain an optimized compound property prediction model, wherein the optimized compound property prediction model is used to predict one property of a compound.

[0074] In some optional implementation manners of the present embodiment, the compound property prediction model training apparatus 500 further includes a fourth obtaining module configured to obtain a target labeled compound dataset, wherein the target labeled compound dataset labels a target property of a compound; and a second fine-tuning module configured to fine-tune the optimized compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model, wherein the fine-tuned compound property prediction model is used to predict the target property of the compound.

[0075] In some optional implementations of this embodiment, the first training module 502 is further configured to: obtain self-supervised information based on the unlabeled compound data in the unlabeled compound dataset; pre-train the graph neural network using the unlabeled compound data as input and the self-supervised information as output to obtain the pre-trained graph neural network.

[0076] In some optional implementations of this embodiment, the self-supervised information includes at least one of the following: local structure of the compound, bond length of the compound, bond angle of the compound, and molecular fingerprint.

[0077] In some optional implementations of this embodiment, the multiple labeled compound datasets include at least two of the following: compound datasets labeled with properties of drug absorption, distribution, metabolism, excretion, and toxicity in ADMET, compound datasets labeled with biological activity, and compound datasets labeled with physicochemical properties of compounds.

[0078] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0079] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0080] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0081] like Figure 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0082] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a magnetic disk, a magneto-optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices over a computer network, such as the Internet, and / or various telecommunication networks.

[0083] The computing unit 601 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs various methods and processes described above, such as the compound property prediction model training method. For example, in some embodiments, the compound property prediction model training method can be implemented as a computer software program, which is tangibly embodied in a machine-readable medium, such as the storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded onto the RAM 603 and executed by the computing unit 601, one or more steps of the compound property prediction model training method described above can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform the compound property prediction model training method by any other appropriate means, such as by means of firmware.

[0084] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0085] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a standalone software package, or entirely on a remote machine or server.

[0086] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0087] To provide for interaction with a user, the systems and techniques described here 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 a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, 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, speech, or tactile input.

[0088] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can 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.

[0089] The computer system can include clients and servers. This relationship can be. The servers are typically remote from the clients with the interactions between them occurring over a communication network. The relationship between a client and a server is one of client-server. The server can be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

[0090] It should be understood that the various forms of flow shown above can be used with reordering, additions, or removals of steps. For example, each of the steps recited in the present disclosure can be performed in parallel, in series, or in a different order, unless otherwise specifically noted, and is not limited to the order recited in this document.

[0091] The specific embodiments described above are not intended to be limiting, and persons skilled in the art will appreciate that various modifications, combinations, sub-combinations and alternatives can be made to the specific embodiments without departing from the spirit and principles of the disclosure. Any further modifications, equivalents, alternatives, and / or improvements made to the specific embodiments described above are intended to fall within the scope of the disclosure.

Claims

1. A method for training a compound property prediction model, comprising: Obtain an unlabeled compound dataset; The graph neural network was pre-trained using the unlabeled compound dataset to obtain the pre-trained graph neural network. Obtain multiple labeled compound datasets, where each labeled compound dataset is labeled with one property of a compound; The pre-trained graph neural network is trained using the multiple labeled compound datasets to obtain a compound property prediction model. The compound property prediction model is used to predict various properties of compounds. The number of training rounds for each task is set to introduce and exit the task. The compound property prediction model is optimized using one of the labeled compound datasets from the plurality of labeled compound datasets to obtain an optimized compound property prediction model, which is used to predict one property of a compound. The step of pre-training the graph neural network using the unlabeled compound dataset to obtain the pre-trained graph neural network includes: Based on the unlabeled compound data in the unlabeled compound dataset, self-supervised information is obtained, which includes at least one of the following: local structure of the compound, bond length of the compound, bond angle of the compound, and molecular fingerprint; Using the unlabeled compound data as input and the self-supervised information as output, the graph neural network is pre-trained to obtain the pre-trained graph neural network.

2. The method according to claim 1, wherein, The method further includes: Obtain a target labeled compound dataset, wherein the target labeled compound dataset labels the target properties of the compounds; The compound property prediction model is fine-tuned using the target labeled compound dataset to obtain a fine-tuned compound property prediction model, wherein the fine-tuned compound property prediction model is used to predict the target properties of compounds.

3. The method according to claim 1, wherein, The method further includes: Obtain a target labeled compound dataset, wherein the target labeled compound dataset labels the target properties of the compounds; The optimized compound property prediction model is fine-tuned using the target labeled compound dataset to obtain a fine-tuned compound property prediction model, wherein the fine-tuned compound property prediction model is used to predict the target properties of compounds.

4. The method according to any one of claims 1-3, wherein, The multiple labeled compound datasets include at least two of the following: compound datasets labeled with properties of drug absorption, distribution, metabolism, excretion, and toxicity in ADMET; compound datasets labeled with biological activity; and compound datasets labeled with physicochemical properties.

5. A compound property prediction model training device, comprising: The first acquisition module is configured to acquire an unlabeled compound dataset; The first training module is configured to pre-train the graph neural network using the unlabeled compound dataset to obtain a pre-trained graph neural network. The second acquisition module is configured to acquire multiple labeled compound datasets, where each labeled compound dataset is labeled with a property of a compound. The second training module is configured to perform multi-task training on the pre-trained graph neural network using the multiple labeled compound datasets to obtain a compound property prediction model. The compound property prediction model is used to predict multiple properties of compounds, and the number of training rounds for each task is set to introduce and exit the task. An optimization module is configured to optimize the compound property prediction model using one of the multiple labeled compound datasets to obtain an optimized compound property prediction model, which is used to predict one property of a compound. The first training module is further configured to: Based on the unlabeled compound data in the unlabeled compound dataset, self-supervised information is obtained, which includes at least one of the following: local structure of the compound, bond length of the compound, bond angle of the compound, and molecular fingerprint; Using the unlabeled compound data as input and the self-supervised information as output, the graph neural network is pre-trained to obtain the pre-trained graph neural network.

6. The apparatus according to claim 5, wherein, The device further includes: The third acquisition module is configured to acquire a target labeled compound dataset, wherein the target labeled compound dataset labels the target properties of the compounds; The first fine-tuning module is configured to fine-tune the compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model, wherein the fine-tuned compound property prediction model is used to predict the target properties of compounds.

7. The apparatus according to claim 6, wherein, The device further includes: The fourth acquisition module is configured to acquire a target labeled compound dataset, wherein the target labeled compound dataset contains the target properties of the labeled compounds; The second fine-tuning module is configured to fine-tune the optimized compound property prediction model using the target labeled compound dataset to obtain a fine-tuned compound property prediction model, wherein the fine-tuned compound property prediction model is used to predict the target properties of compounds.

8. The apparatus according to any one of claims 5-7, wherein, The multiple labeled compound datasets include at least two of the following: compound datasets labeled with properties of drug absorption, distribution, metabolism, excretion, and toxicity in ADMET; compound datasets labeled with biological activity; and compound datasets labeled with physicochemical properties.

9. 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 to enable the at least one processor to perform the method of any one of claims 1-4.

10. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-4.

11. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-4.

Images