Small molecule attribute prediction model construction method and device, equipment and storage medium

Abstract

The application discloses a small molecule attribute prediction model construction method and device, equipment and a storage medium, relates to the computer technical field, and is used for improving the effect and accuracy of predicting drug-target interaction. Including: based on the parameters corresponding to the three-layer GCN network included in the pre-training model, a target model is constructed, and the target model includes multiple parallel two-layer full connection layers; parameter information corresponding to multiple compounds is acquired, and the parameter information corresponding to the multiple compounds is input into the target model in sequence to obtain multiple target results corresponding to each compound; based on the parameter information corresponding to each compound in the multiple compounds and the multiple target results corresponding to each compound, the parameters of the third layer GCN network of the three-layer GCN network included in the target model are adjusted to obtain a small molecule attribute prediction model. The application is applied to the scene of predicting drug-target interaction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, and particularly relates to a small molecule attribute prediction model construction method and device, equipment and a storage medium. BACKGROUND

[0002] Drug development is a complex, long and expensive process, usually requiring 10-15 years of research and 8-15 billion dollars of funds to change a drug from an abstract concept to a marketable product. Each year, 90% of drugs fail in the evaluation of the Food and Drug Administration (FDA), making them unable to be used for actual treatment. In November 2018, a study estimated the total cost of trials of new drugs approved by the FDA, and the study showed that the average trial cost of 59 new drugs approved by the FDA during 2015-2016 was 19 million US dollars. Therefore, the development of a computational method to identify potential drug-target (target point) interactions with the minimum error rate is increasingly sought after. Most of the methods developed so far are to use binary classification to predict whether there is an interaction between a drug and its target point.

[0003] However, traditional machine learning methods are usually used to predict drug-target interactions. The traditional machine learning model is trained on an internal dataset established for a specific purpose, and its accuracy is limited, resulting in poor effect and low accuracy in predicting drug-target interactions. SUMMARY

[0004] The present application provides a small molecule attribute prediction model construction method and device, equipment and a storage medium, which are used to improve the effect and accuracy of predicting drug-target interactions.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] In a first aspect, a small molecule attribute prediction model construction method is provided. The method includes: constructing a target model based on parameters corresponding to a three-layer GCN network included in a pre-trained model, the target model including a plurality of parallel two-layer fully connected layers; obtaining parameter information corresponding to a plurality of compounds and inputting the parameter information corresponding to the plurality of compounds into the target model in sequence to obtain a plurality of target results corresponding to each compound, one of the plurality of parallel two-layer fully connected layers corresponding to one target result, the parameter information including at least one of the following parameters: p53 transcriptional activity level, HIF-1 transcriptional activity level, Wnt signal transduction pathway, PPARγ transcriptional activity level, and STAT3 / NF-κB transcriptional activity level, the number of the plurality of parallel two-layer fully connected layers being the same as the number of parameters included in the parameter information; and adjusting parameters of a third GCN network of the three-layer GCN network included in the target model based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each compound to obtain a small molecule attribute prediction model.

[0007] In a possible implementation, the method further includes: obtaining a plurality of preset training data from a preset database, each of the plurality of preset training data including a molecular formula and a preset logP value; inputting each of the plurality of preset training data into the preset base model in sequence to obtain a target logP value corresponding to each preset training data; and adjusting parameters of a three-layer GCN network included in the base model based on the target logP value corresponding to each preset training data and the preset logP value to obtain the pre-trained model.

[0008] In a possible implementation, inputting each of the plurality of preset training data into the preset base model in sequence to obtain a target logP value corresponding to each preset training data includes: inputting each of the plurality of preset training data into the preset base model in sequence to obtain a feature vector corresponding to each preset training data based on the three-layer GCN network included in the base model; and obtaining the target logP value corresponding to each preset training data based on the feature vector corresponding to each preset training data and a two-layer fully connected neural network layer included in the base model.

[0009] In a possible implementation, constructing a target model based on parameters corresponding to a three-layer GCN network included in a pre-trained model includes: replacing a two-layer fully connected neural network layer included in the pre-trained model with a plurality of parallel two-layer fully connected neural network layers, and locking parameters of the first two GCN networks of the three-layer GCN network included in the pre-trained model to obtain the target model.

[0010] In a second aspect, a small molecule attribute prediction model construction device is provided. The small molecule attribute prediction model construction device includes a processing unit and an acquisition unit. The processing unit is configured to construct a target model based on parameters corresponding to a three-layer GCN network included in a pre-trained model. The target model includes a plurality of parallel two-layer fully connected layers. The acquisition unit is configured to acquire parameter information corresponding to a plurality of compounds. The processing unit is configured to sequentially input the parameter information corresponding to the plurality of compounds into the target model to obtain a plurality of target results corresponding to each compound. One of the plurality of parallel two-layer fully connected layers corresponds to one target result. The parameter information includes at least one of the following parameters: a p53 transcriptional activity level, a HIF-1 transcriptional activity level, a Wnt signal transduction pathway, a PPARγ transcriptional activity level, and a STAT3 / NF-κB transcriptional activity level. The number of the plurality of parallel two-layer fully connected layers is the same as the number of parameters included in the parameter information. The processing unit is configured to adjust parameters of a third GCN network of the three-layer GCN network included in the target model based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each of the plurality of compounds to obtain a small molecule attribute prediction model.

[0011] In a possible implementation, the acquisition unit is configured to acquire a plurality of preset training data from a preset database. Each of the plurality of preset training data includes a molecular formula and a preset logP value. The processing unit is configured to sequentially input each of the plurality of preset training data into a preset base model to obtain a target logP value corresponding to each of the preset training data. The processing unit is configured to adjust parameters of a three-layer GCN network included in the base model based on the target logP value corresponding to each of the preset training data and the preset logP value to obtain the pre-trained model.

[0012] In a possible implementation, the processing unit is configured to sequentially input each of the plurality of preset training data into the preset base model to obtain a feature vector corresponding to each of the preset training data based on the three-layer GCN network included in the base model. The processing unit is configured to obtain the target logP value corresponding to each of the preset training data based on the feature vector corresponding to each of the preset training data and a two-layer fully connected neural network layer included in the base model.

[0013] In a possible implementation, the processing unit is configured to replace one two-layer fully connected neural network layer included in the pre-trained model with a plurality of parallel two-layer fully connected neural network layers, and lock parameters of the first two GCN networks of the three-layer GCN network included in the pre-trained model to obtain the target model.

[0014] In a third aspect, an electronic device includes a processor and a memory. The memory is configured to store one or more programs including computer-executable instructions. When the electronic device is running, the processor executes the computer-executable instructions stored in the memory, so that the electronic device executes the small-molecule attribute prediction model construction method according to the first aspect.

[0015] In a fourth aspect, a computer-readable storage medium storing one or more programs is provided. The one or more programs include instructions that, when executed by a computer, cause the computer to perform the small-molecule attribute prediction model construction method according to the first aspect.

[0016] The present application provides a small-molecule attribute prediction model construction method, device, equipment and storage medium, which is applied to the scene of predicting drug-target interaction. First, based on the parameters corresponding to the three-layer GCN network included in the pre-trained model, a target model including a plurality of parallel two-layer fully connected layers is constructed. Then, further, the parameter information corresponding to a plurality of compounds is obtained, and the parameter information corresponding to the plurality of compounds is input into the target model in turn, so as to obtain a plurality of target results corresponding to each compound. Thus, based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each compound, the parameters of the third GCN network of the three-layer GCN network included in the target model are adjusted, so as to obtain a small-molecule attribute prediction model. Through the above method, based on the obtained parameter information corresponding to the plurality of compounds, the target model constructed based on the parameters corresponding to the three-layer GCN network included in the pre-trained model is trained, so as to adjust the parameters of the third GCN network of the three-layer GCN network included in the target model based on the training result, and obtain a small-molecule attribute prediction model. Thus, through the obtained small-molecule attribute prediction model, the effect and accuracy of predicting drug-target interaction can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A small-molecule attribute prediction model construction system structure schematic diagram is provided for the embodiments of the present application;

[0018] Figure 2 A small-molecule attribute prediction model construction method flowchart is provided for the embodiments of the present application Figure 1 ;

[0019] Figure 3 A small-molecule attribute prediction model construction method flowchart is provided for the embodiments of the present application Figure 2 ;

[0020] Figure 4 A model schematic diagram is provided for the embodiments of the present application;

[0021] Figure 5 A small molecule attribute prediction model construction method flowchart provided for an embodiment of the present application Figure 3 ;

[0022] Figure 6 A small molecule attribute prediction model construction method flowchart provided for an embodiment of the present application Figure 4 ;

[0023] Figure 7 A small molecule attribute prediction model construction device structure diagram provided for an embodiment of the present application

[0024] Figure 8 An electronic device structure diagram provided for an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

[0026] In the description of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this document is only a description of the association relationship of the associated objects, which means that there can be three kinds of relationships, for example, A and / or B, which can mean: A exists alone, A and B exist together, and B exists alone. In addition, "at least one" "multiple" means two or more. "First", "second", etc. do not limit the quantity and execution order, and "first", "second", etc. do not necessarily mean different.

[0027] Developing computational methods to identify potential drug-target interactions with minimal error rates is increasingly popular. Most of the methods developed so far are using binary classification to predict whether there is an interaction between a drug and its target. However, it is more valuable and challenging to predict the binding strength between a drug and its target. Such drug-target interactions can have no effect if the binding strength is not enough. Therefore, it is of great value to develop methods to predict the binding affinity of drugs and targets (Drug-Target Affinity, DTA).

[0028] The small molecule attribute prediction model construction method provided in the embodiments of the present application can be applied to a small molecule attribute prediction model construction system. Figure 1 An structure diagram of the small molecule attribute prediction model construction system is shown. As shown in Figure 1 The small molecule attribute prediction model construction system 20 includes an electronic device 21 and a server 22.

[0029] The electronic device 21 is a device for performing a small molecule attribute prediction model construction method, and is configured to construct a target model including a plurality of parallel two-layer fully connected layers based on parameters corresponding to a three-layer GCN network included in a pre-trained model; then acquire parameter information corresponding to a plurality of compounds, and sequentially input the parameter information corresponding to the plurality of compounds into the target model to obtain a plurality of target results corresponding to each compound; and adjust parameters of a third GCN network of the three-layer GCN network included in the target model based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each compound, to obtain a small molecule attribute prediction model.

[0030] The server 22 is a server corresponding to the electronic device 21 for performing the small molecule attribute prediction model construction method, and is configured to provide network connection for the electronic device 21 and realize data interaction between the electronic device 21, and provide data information for the electronic device 21.

[0031] A small molecule attribute prediction model construction method provided by an embodiment of the present application will be described below with reference to the accompanying drawings. As shown in Figure 2 A small molecule attribute prediction model construction method provided by an embodiment of the present application includes S201-S203:

[0032] S201, constructing a target model based on parameters corresponding to a three-layer GCN network included in a pre-trained model.

[0033] The target model includes a plurality of parallel two-layer fully connected layers.

[0034] The small molecule attribute prediction model construction method provided by the embodiment of the present application discloses a model for predicting drug attributes based on a graph convolutional neural network, which is used to predict the attributes of marine small molecules, and uses deep learning and a transfer learning strategy for molecular attribute prediction.

[0035] It should be noted that the model construction method provided by the embodiment of the present application is divided into two parts, the first part is a pre-trained model (i.e., a trained model is constructed), and the second part is a multi-task learning part (i.e., a small molecule attribute prediction model is constructed).

[0036] First, the multi-task learning part of the second part will be described in detail.

[0037] Optionally, after obtaining the pre-trained model (Pre-trained Model) in advance, a target model (Model) can be constructed based on parameters corresponding to a three-layer GCN network included in the pre-trained model, so as to train a small molecule attribute prediction model based on the target model.

[0038] Specifically, based on the parameters corresponding to the three-layer GCN network included in the pre-trained model, the target model is constructed by migrating the pre-trained model to a multi-task model for predicting multiple molecular attributes, replacing the second fully connected layer of the pre-trained model with multiple parallel second fully connected layers, and obtaining the target model, so as to output the prediction values of multiple molecular properties.

[0039] S202, obtain parameter information corresponding to a plurality of compounds, and input the parameter information corresponding to the plurality of compounds into the target model in turn to obtain a plurality of target results corresponding to each compound.

[0040] Among them, one of the multiple parallel second fully connected layers corresponds to one target result, and the parameter information includes at least one of the following parameters: p53 transcriptional activity level, HIF-1 transcriptional activity level, Wnt signal transduction pathway, PPARγ transcriptional activity level, STAT3 / NF-κB transcriptional activity level, and the number of multiple parallel second fully connected layers is the same as the number of parameters included in the parameter information.

[0041] Optionally, the parameter information corresponding to a plurality of compounds (for example, 18834 compounds) is obtained from the database, and the luciferase activity of the plurality of compounds is detected in the above-mentioned five screening systems (i.e., p53 transcriptional activity level, HIF-1 transcriptional activity level, Wnt signal transduction pathway, PPARγ transcriptional activity level, STAT3 / NF-κB transcriptional activity level), and compared with the control group without compound treatment, to obtain the activation or inhibition ratio of luciferase activity of each compound in the five high-throughput screening systems, to characterize the effect of the compound on the signal pathway involved.

[0042] Further, the small molecule compounds to be tested are divided into a plurality of data sets (for example, 8 data sets), and the number of molecules in each database and the value range of each task are as shown in Table 1.

[0043] Among them, parameter 1 is used to indicate the parameter interval of the p53 transcriptional activity level corresponding to the plurality of compounds included in each data set, parameter 2 is used to indicate the parameter interval of the HIF-1 transcriptional activity level corresponding to the plurality of compounds included in each data set, parameter 3 is used to indicate the parameter interval of the Wnt signal transduction pathway corresponding to the plurality of compounds included in each data set, parameter 4 is used to indicate the parameter interval of the PPARγ transcriptional activity level corresponding to the plurality of compounds included in each data set, and parameter 5 is used to indicate the parameter interval of the STAT3 / NF-κB transcriptional activity level corresponding to the plurality of compounds included in each data set.

[0044] Table 1

[0045] Dataset Molecular weight Parameter 1 Parameter 2 Parameter 3 Parameter 4 Parameter 5 #1 439 0-5.67 0-2.65 0-2.80 0-1.56 0-1.75 #2 514 0-5.77 0-6.53 0-64.17 0-15.17 0-14.71 #3 830 0-3.48 0-8.49 0-16.57 0-35.23 0-34.05 #4 1813 0-4.19 0-9.36 0-54.14 0-21.64 0-24.80 #5 3187 0-2.39 0-3.26 0-3.27 0.77-1.66 0.02-7.39 #6 3520 0-4.41 0-5.23 0-23.57 0-8.27 0-32.98 #7 4131 0-3.42 0.02-2.19 0-5.33 0-27.35 0-33.61 #8 4400 0-6.73 0-8.42 0-41.66 0-17.38 0-13.77

[0046] S203, based on the parameter information corresponding to each compound in the plurality of compounds and the plurality of target results corresponding to each compound, adjusting the parameters of the third layer GCN network of the three-layer GCN network included in the target model to obtain a small molecule attribute prediction model.

[0047] In one design, as shown in Figure 3 In the small molecule attribute prediction model construction method provided by the embodiments of the present application, the method in step S201 can specifically include step S2011:

[0048] S2011, replace one two-layer fully connected neural network layer included in the pre-trained model with a plurality of parallel two-layer fully connected neural network layers, and lock the first two layers of the three-layer GCN network included in the pre-trained model to obtain a target model.

[0049] Optionally, as shown in Figure 4 After the target model is constructed based on the parameters corresponding to the three-layer GCN network included in the pre-trained model, the parameters of the first two layers of the three-layer GCN network included in the target model need to be locked, so that only the parameters of the third layer GCN network of the three-layer GCN network included in the target model are adjusted in subsequent training.

[0050] In the embodiments of the present application, the pre-trained model is migrated to the model (i.e. target model) for predicting the attributes of marine small molecules, and multiple attribute values are predicted at the same time. In the process of training the multi-task model, the parameters of the first two layers of the graph convolutional neural network are frozen, and only the third layer of the graph convolutional neural network is retrained. Then the extracted feature vectors are input into multiple three-layer fully connected networks for predicting the attribute values of marine small molecules.

[0051] In one design, as shown in Figure 5 The small molecule attribute prediction model construction method provided by the embodiments of the present application can further include S301-S303:

[0052] S301, obtaining a plurality of preset training data from a preset database.

[0053] Each of the plurality of preset training data includes: molecular formula, preset logP value.

[0054] Optionally, the preset database can be a DrugBank database.

[0055] Further, the first part of the pre-trained model (i.e. the part of the training model) is described in detail.

[0056] Specifically, the pre-trained model uses the DrugBank database to pre-train the graph convolutional neural network. The training result is the predicted logP value (i.e. the target logP value). The feature vector and molecular graph are input into the graph convolutional neural network to obtain the feature vector of the molecule, which is then input into a three-layer fully connected network.

[0057] It's important to note that logP is a widely used descriptor. A compound's logP is used in many different quantitative structure-activity relationship (QSAR) analyses to predict a wide range of activities and properties. It has been used for skin penetration to determine drug similarity. It has also been used to estimate solubility and is considered important for predicting general toxicity in aquatic animals. One method for predicting logP involves calculating the molecule's descriptors and then using these descriptors to predict logP. The complexity of these descriptors varies, ranging from simple carbon and heteroatom counting to methods that attempt to correlate the molecule's solubility energy with complex calculations to predict its log value.

[0058] Table 2

[0059] Drug classification Number Small molecule drugs 11923 Biotech drugs 2687 Approved drugs 4237 Approved small molecule drugs 2721 Nutriceutical drugs 132 Experimental drugs 6665 Illegal drugs 205 Withdrawn drugs 269

[0060] For example, as shown in Table 2, the DrugBank database is a comprehensive, freely accessible online database containing information on drugs and drug targets. Currently, the DrugBank database contains 11,923 small molecule drugs, including 2,687 biotechnology drugs, 4,237 approved drugs, 2,721 approved small molecule drugs, 132 health supplements, 6,665 investigational drugs, 205 illicit drugs, and 269 withdrawn drugs.

[0061] The DrugBank database contains 19,028 targets, including 4,902 unique targets, 9,303 approved drug targets, 1,024 health supplement targets, and 8,197 investigational targets. 8,266 drug information entries were parsed from the database. The logP value was calculated using the ALOGPS algorithm, and the logP value ranges from -4.2 to 10.25.

[0062] S302. Input each preset training data from multiple preset training data into the preset base model in sequence to obtain the target logP value corresponding to each preset training data.

[0063] Optionally, the target logP value corresponding to each preset training data can be obtained by sequentially inputting each preset training data into a preset base model.

[0064] S303, based on the target logP value corresponding to each preset training data and the preset logP value, adjusting the parameters of the three-layer GCN network included in the basic model to obtain a pre-trained model.

[0065] Further, based on the size relationship between the target logP value corresponding to each preset training data and the preset logP value, the parameters of the three-layer GCN network included in the basic model are adjusted, and after all the preset training data are input, the pre-trained model can be obtained.

[0066] It can be understood that based on the above method, the transfer learning technology can be used to predict the small molecule attribute, and the result is compared with the result of the non-transfer learning method to outline whether it can provide at least comparable accuracy. Since it is necessary to predict the p53 transcriptional activity level, the HIF-1 transcriptional activity level, the Wnt signal transduction pathway, the PPARγ transcriptional activity level, and the STAT3 / NF-κB transcriptional activity level, and to obtain the relationship between the five attributes, the feature vector obtained by the upper graph convolution is input into five parallel fully connected networks to predict the five attributes. The five tasks can transmit information to each other and optimize each other to obtain the best prediction result.

[0067] In order to better initialize the numerical value of the model, the transfer learning method is adopted. First, the model is pre-trained in the DrugBank dataset. During pre-training, the supervised learning method is used to convert the molecule into a molecular graph, and then a three-layer GCN network is used as a feature extractor to obtain the feature vector of the molecule, which is then input into a fully connected neural network to fit the logP attribute of the molecule. A pre-trained model is obtained, and then the pre-trained model is migrated to the multi-task model for predicting the above five molecular attributes. By replacing the fully connected layer of the pre-trained model with five parallel two-layer fully connected layers, the prediction values of the five molecular properties are output. During training, the first two layers of the GCN are frozen, and the third layer of the GCN is fine-tuned to obtain the best fitting effect of the model.

[0068] In one design, as shown in FIG. Figure 6 The method in step S302 can specifically include steps S401-S402 in the small molecule attribute prediction model construction method provided by the embodiments of the present application.

[0069] S401, each of the plurality of preset training data is sequentially input into a preset basic model, and a three-layer GCN network included in the basic model is used to obtain a feature vector corresponding to each preset training data.

[0070] S402, obtaining a target logP value corresponding to each preset training data based on a feature vector corresponding to each preset training data and a two-layer fully connected neural network layer included in the basic model.

[0071] Optionally, in the process of training the pre-trained model, each molecule in each preset training data is converted into a molecular graph by using a supervised learning method, and then a three-layer GCN network is used as a feature extractor to obtain a feature vector of the molecule, and then input into a two-layer fully connected neural network layer to fit the logP attribute of the molecule, and then adjust the parameters of the three-layer GCN network included in the basic model to obtain the pre-trained model.

[0072] In the embodiments of the present application, by using the migration learning strategy, the three-layer graph convolutional neural network and the fully connected neural network, the p53 transcriptional activity level, the HIF-1 transcriptional activity level, the Wnt signal transduction pathway, the PPARγ transcriptional activity level and the STAT3 / NF-κB transcriptional activity level can be simultaneously predicted.

[0073] The present application provides a small molecule attribute prediction model construction method, which first constructs a target model including a plurality of parallel two-layer fully connected layers based on the parameters corresponding to the three-layer GCN network included in the pre-trained model. Then further obtain the parameter information corresponding to a plurality of compounds, and sequentially input the parameter information corresponding to a plurality of compounds into the target model, thereby obtaining a plurality of target results corresponding to each compound. Therefore, based on the parameter information corresponding to each compound in the plurality of compounds and the plurality of target results corresponding to each compound, the parameters of the third layer GCN network of the three-layer GCN network included in the target model are adjusted, and a small molecule attribute prediction model can be obtained. Through the above method, based on the obtained parameter information corresponding to a plurality of compounds, the target model constructed based on the parameters corresponding to the three-layer GCN network included in the pre-trained model is trained, and then based on the training result, the parameters of the third layer GCN network of the three-layer GCN network included in the target model are adjusted, that is, a small molecule attribute prediction model is obtained. Therefore, through the obtained small molecule attribute prediction model, the effect and accuracy of predicting drug-target interaction can be improved.

[0074] The above describes the scheme provided by the embodiments of the present application from the perspective of the method. To implement the above functions, hardware structures and / or software modules corresponding to the functions are included. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of the examples described in the embodiments disclosed herein, the embodiments of the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is implemented in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0075] The embodiments of the present application can divide a small molecule attribute prediction model construction device into functional modules according to the above method examples. For example, each functional module can be divided according to each function, or two or more functions can be integrated in one processing module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. Optionally, the division of the modules in the embodiments of the present application is illustrative, and is only a logical functional division. In actual implementation, another division method can be used.

[0076] Figure 7 A structural diagram of a small molecule attribute prediction model construction device provided by the embodiments of the present application is shown in FIG. 1. A small molecule attribute prediction model construction device 100 is used to improve the effect and accuracy of predicting drug-target interactions, for example, to implement a small molecule attribute prediction model construction method shown in FIG. 2. The small molecule attribute prediction model construction device 100 includes a processing unit 1001 and an acquisition unit 1002. Figure 7 Figure 2 The processing unit 1001 is configured to construct a target model based on parameters corresponding to a three-layer GCN network included in a pre-training model. The target model includes a plurality of parallel two-layer fully connected layers.

[0077] The acquisition unit 1002 is configured to acquire parameter information corresponding to a plurality of compounds.

[0078] The processing unit 1001 is configured to sequentially input the parameter information corresponding to the plurality of compounds into the target model to obtain a plurality of target results corresponding to each compound. One of the plurality of parallel two-layer fully connected layers corresponds to one target result. The parameter information includes at least one of the following parameters: a p53 transcriptional activity level, a HIF-1 transcriptional activity level, a Wnt signal transduction pathway, a PPARγ transcriptional activity level, and a STAT3 / NF-κB transcriptional activity level. The number of the plurality of parallel two-layer fully connected layers is the same as the number of parameters included in the parameter information.

[0079] The processing unit 1001 is configured to sequentially input the parameter information corresponding to the plurality of compounds into the target model to obtain a plurality of target results corresponding to each compound. One of the plurality of parallel two-layer fully connected layers corresponds to one target result. The parameter information includes at least one of the following parameters: a p53 transcriptional activity level, a HIF-1 transcriptional activity level, a Wnt signal transduction pathway, a PPARγ transcriptional activity level, and a STAT3 / NF-κB transcriptional activity level. The number of the plurality of parallel two-layer fully connected layers is the same as the number of parameters included in the parameter information.​

[0080] The processing unit 1001 is configured to adjust parameters of a third GCN network of a three-layer GCN network included in the target model based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each of the plurality of compounds, to obtain the small-molecule attribute prediction model.

[0081] In a possible implementation, in the small-molecule attribute prediction model construction apparatus 100 provided by the embodiment of the present application, the obtaining unit 1002 is configured to obtain a plurality of preset training data from a preset database, each of the plurality of preset training data including a molecular formula and a preset logP value.

[0082] The processing unit 1001 is configured to sequentially input each of the plurality of preset training data into a preset base model to obtain a target logP value corresponding to each of the preset training data.

[0083] The processing unit 1001 is configured to adjust parameters of a three-layer GCN network included in the base model based on the target logP value corresponding to each of the preset training data and the preset logP value, to obtain a pre-training model.

[0084] In a possible implementation, in the small-molecule attribute prediction model construction apparatus 100 provided by the embodiment of the present application, the processing unit 1001 is configured to sequentially input each of the plurality of preset training data into a preset base model, and obtain a feature vector corresponding to each of the preset training data based on a three-layer GCN network included in the base model.

[0085] The processing unit 1001 is configured to obtain a target logP value corresponding to each of the preset training data based on the feature vector corresponding to each of the preset training data and a two-layer fully connected neural network layer included in the base model.

[0086] In a possible implementation, in the small-molecule attribute prediction model construction apparatus 100 provided by the embodiment of the present application, the processing unit 1001 is configured to replace a two-layer fully connected neural network layer included in the pre-training model with a plurality of parallel two-layer fully connected neural network layers, and lock parameters of the first two GCN networks of the three-layer GCN network included in the pre-training model, to obtain a target model.

[0087] In the case of implementing the functions of the above-mentioned integrated modules in the form of hardware, the embodiment of the present application provides another possible structural diagram of the electronic device involved in the above-mentioned embodiments. As shown in FIG. 9, an electronic device 90 is configured to improve the effect and accuracy of predicting drug-target interactions, for example, to perform the method 1000. Figure 8 Figure 2 ​An embodiment of a small molecule property prediction model construction method is shown. The electronic device 90 includes a processor 901, a memory 902, and a bus 903. The processor 901 and the memory 902 can be connected through the bus 903.

[0088] The processor 901 is a control center of the communication device, and can be one processor or a general term of multiple processing elements. For example, the processor 901 can be a general central processing unit (CPU), or other general-purpose processors, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0089] As an embodiment, the processor 901 can include one or more CPUs, such as the CPU 0 and the CPU 1 shown in FIG. 1. Figure 8

[0090] The memory 902 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to this.

[0091] As a possible implementation, the memory 902 can exist independently of the processor 901, and the memory 902 can be connected to the processor 901 through the bus 903, for storing instructions or program code. When the processor 901 invokes and executes the instructions or program code stored in the memory 902, the small molecule property prediction model construction method provided in the embodiments of the present application can be implemented.

[0092] In another possible implementation, the memory 902 can also be integrated with the processor 901.

[0093] ​The bus 903 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, Figure 8 Only one thick line is used to represent the bus in the figure, but it does not mean that there is only one bus or only one type of bus.

[0094] It should be noted that, Figure 8 The structure shown does not constitute a limitation on the electronic device 90. In addition to Figure 8 the illustrated components, the electronic device 90 can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0095] As an example, in combination with Figure 7 , the functions implemented by the processing unit 1001 and the acquisition unit 1002 in the electronic device are the same as those of the processor 901 in Figure 8 .

[0096] Optionally, as Figure 8 indicated, the electronic device 90 provided by the embodiments of the present application can further include a communication interface 904.

[0097] The communication interface 904 is configured to connect with other devices through a communication network. The communication network can be an Ethernet, a wireless access network, a wireless local area network (WLAN), or the like. The communication interface 904 can include a receiving unit for receiving data, and a sending unit for sending data.

[0098] In one design, in the electronic device provided by the embodiments of the present application, the communication interface can also be integrated in the processor.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units is taken as an example. In actual application, the above-mentioned functions can be completed by different functional units according to needs, that is, the internal structure of the device is divided into different functional units to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0100] The embodiment of the present application further provides a computer readable storage medium, and the computer readable storage medium stores instructions. When a computer executes the instructions, the computer executes each step in the method flow shown in the method embodiment.

[0101] The embodiment of the present application provides a computer program product containing instructions, which, when executed on a computer, cause the computer to execute a small molecule attribute prediction model construction method in the method embodiment.

[0102] The computer readable storage medium may, for example, be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any combination of the above. More specific examples (a non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a register, a hard disk, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any other suitable combination of the above, or any other physical medium that can be used to carry or store desired computer readable instructions in a form accessible by a processing system, such as a computer.

[0103] An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. Of course, the storage medium can be a part of the processor. The processor and the storage medium can be located in an Application Specific Integrated Circuit (ASIC).

[0104] In the embodiment of the present application, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or apparatus.

[0105] Since the electronic device, the computer readable storage medium and the computer program product in the embodiment of the present application can be applied to the above method, the technical effects that can be obtained are also referable to the above method embodiment, and the embodiment of the present application will not be described here.

[0106] The above is merely specific implementation of the present application, but the protection scope of the present application is not limited to this. Any change or replacement within the technical scope disclosed in the present application should be covered in the protection scope of the present application.

Claims

1. A method for constructing a small molecule property prediction model, characterized by, The method comprises: obtaining a plurality of preset training data from a preset database, each of the plurality of preset training data comprising: a molecular formula, a preset logP value; inputting each of the plurality of preset training data into a preset base model in sequence to obtain a target logP value corresponding to each of the preset training data; adjusting parameters of a three-layer GCN network included in the base model based on the target logP value corresponding to each of the preset training data and the preset logP value to obtain a pre-training model; replacing a two-layer fully connected neural network layer included in the pre-training model with a plurality of parallel two-layer fully connected neural network layers, and locking parameters of the first two layers of the three-layer GCN network included in the pre-training model to obtain a target model; the target model comprises a plurality of parallel two-layer fully connected layers; obtaining parameter information corresponding to a plurality of compounds, and inputting the parameter information corresponding to the plurality of compounds into the target model in sequence to obtain a plurality of target results corresponding to each of the compounds, one of the plurality of parallel two-layer fully connected layers corresponding to one target result, the parameter information comprising at least one of the following parameters: p53 transcriptional activity level, HIF-1 transcriptional activity level, Wnt signal transduction pathway, PPAR γ transcriptional activity level, STAT3 / NF-κB transcriptional activity level, the number of the plurality of parallel two-layer fully connected layers being the same as the number of parameters included in the parameter information; adjusting parameters of the third layer of the three-layer GCN network included in the target model based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each of the compounds to obtain a small molecule attribute prediction model.

2. The method of claim 1, wherein, The method comprises: inputting each of the plurality of preset training data into a preset base model in sequence to obtain a target logP value corresponding to each of the preset training data, comprising: inputting each of the plurality of preset training data into a preset base model in sequence to obtain a feature vector corresponding to each of the preset training data based on a three-layer GCN network included in the base model; 3. A small molecule property prediction model construction apparatus characterized by comprising: obtaining a target logP value corresponding to each of the preset training data based on the feature vector corresponding to each of the preset training data and a two-layer fully connected neural network layer included in the base model. The small molecule attribute prediction model construction device comprises a processing unit and an obtaining unit. The obtaining unit is configured to obtain a plurality of preset training data from a preset database, each of the plurality of preset training data comprising: a molecular formula, a preset logP value; The processing unit is configured to input each of the plurality of preset training data into a preset base model in sequence to obtain a target logP value corresponding to each of the preset training data; The processing unit is configured to adjust parameters of a three-layer GCN network included in the base model based on the target logP value corresponding to each of the preset training data and the preset logP value to obtain a pre-training model; The processing unit is configured to replace a two-layer fully connected neural network layer included in the pre-training model with a plurality of parallel two-layer fully connected neural network layers, and lock parameters of the first two layers of the three-layer GCN network included in the pre-training model to obtain a target model; the target model comprises a plurality of parallel two-layer fully connected layers. The processing unit is configured to replace a two-layer fully connected neural network layer included in the pre-trained model with a plurality of parallel two-layer fully connected neural network layers, and lock parameters of the first two layers of GCN network of the three-layer GCN network included in the pre-trained model, to obtain a target model; the target model includes a plurality of parallel two-layer fully connected layers; The acquisition unit is configured to acquire parameter information corresponding to a plurality of compounds; The processing unit is configured to sequentially input the parameter information corresponding to the plurality of compounds into the target model, to obtain a plurality of target results corresponding to each compound, one parallel two-layer fully connected layer in the plurality of parallel two-layer fully connected layers corresponding to one target result, the parameter information including at least one of the following parameters: p53 transcriptional activity level, HIF-1 transcriptional activity level, Wnt signal transduction pathway, PPAR γ transcriptional activity level, and STAT3 / NF-κB transcriptional activity level, the number of the plurality of parallel two-layer fully connected layers being the same as the number of the parameters included in the parameter information; The processing unit is configured to adjust parameters of the third layer of GCN network of the three-layer GCN network included in the target model based on the parameter information corresponding to each of the plurality of compounds and the plurality of target results corresponding to each of the plurality of compounds, to obtain a small molecule attribute prediction model.

4. The small molecule property prediction model construction apparatus according to claim 3, wherein The processing unit is configured to sequentially input each of the plurality of preset training data into a preset basic model, and obtain a feature vector corresponding to each preset training data based on a three-layer GCN network included in the basic model; The processing unit is configured to obtain a target logP value corresponding to each preset training data based on the feature vector corresponding to each preset training data and a two-layer fully connected neural network layer included in the basic model.

5. An electronic device, comprising: The processing unit is configured to obtain a target logP value corresponding to each preset training data based on the feature vector corresponding to each preset training data and a two-layer fully connected neural network layer included in the basic model. The processor and the memory; wherein the memory is configured to store one or more programs, the one or more programs including computer execution instructions, when the electronic device is running, the processor executes the computer execution instructions stored in the memory, so that the electronic device executes the small molecule attribute prediction model construction method in any one of claims 1-2.

6. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method comprising: The one or more programs include instructions that, when executed by a computer, cause the computer to perform the small molecule attribute prediction model construction method of any one of claims 1-2.

Images