Method, device, equipment and storage medium for determining compound property parameters

Abstract

The present application provides a method, apparatus, device, and storage medium for determining compound attribute parameters, belonging to the field of computer technology. The method includes: training a prediction model based on at least one first sample data, the first sample data including the structural data and sample partition coefficient of a first sample compound, the sample partition coefficient being the partition coefficient of the first sample compound between the aqueous phase and the lipid phase, the trained prediction model being used to determine the partition coefficient of any compound based on the structural data of any compound; retraining the trained prediction model based on at least one second sample data, the second sample data including the structural data and sample attribute parameters of a second sample compound, the sample attribute parameters being attribute parameters related to the partition coefficient; calling the retrained prediction model to determine the target attribute parameters of any compound based on the structural data of any compound. The above method can improve the accuracy of determining compound attribute parameters.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for determining compound property parameters. Background Art

[0002] Compounds possess various property parameters, such as solubility and permeability, which are crucial for compound research. For example, solubility can be used to study the compound's migration behavior within organisms and the environment, and is a crucial parameter in many biological and physicochemical processes. Therefore, determining these property parameters is often necessary when studying compounds.

[0003] In related technologies, a prediction model is typically trained using the property parameters of sample compounds, and then invoked to determine the property parameters of any compound. For example, a prediction model is trained using the solubility of a compound, and then invoked to determine the solubility of any compound. However, when a compound has relatively few property parameters, the data available for model training is very limited, resulting in low prediction model accuracy. Summary of the Invention

[0004] The present invention provides a method, apparatus, device, and storage medium for determining compound property parameters, which can improve the accuracy of determining compound property parameters. The technical solution is as follows:

[0005] In one aspect, a method for determining a compound property parameter is provided, the method comprising:

[0006] Training a prediction model based on at least one first sample data, wherein the first sample data includes structural data and a sample partition coefficient of a first sample compound, wherein the sample partition coefficient is a partition coefficient of the first sample compound between an aqueous phase and a lipid phase, and the trained prediction model is used to determine the partition coefficient of any compound based on the structural data of the compound;

[0007] retraining the trained prediction model according to at least one second sample data, wherein the second sample data includes structural data and sample attribute parameters of a second sample compound, wherein the sample attribute parameters are attribute parameters related to the partition coefficient;

[0008] The retrained prediction model is called to determine the target attribute parameters of any compound according to the structural data of any compound.

[0009] In a possible implementation, retraining the trained prediction model according to at least one second sample data includes:

[0010] For any second sample data, calling the trained prediction model, and determining the predicted attribute parameter of the second sample compound according to the structural data of the second sample compound in the second sample data;

[0011] The trained prediction model is retrained according to the prediction attribute parameters and the sample attribute parameters in the second sample data.

[0012] In another possible implementation, retraining the trained prediction model according to the prediction attribute parameter and the sample attribute parameter in the second sample data includes:

[0013] Determining a first loss value of the trained prediction model according to the prediction attribute parameter and the sample attribute parameter in the second sample data;

[0014] The trained prediction model is retrained according to the first loss value.

[0015] In another possible implementation, the trained prediction model includes multiple network layers, and retraining the trained prediction model according to at least one second sample data includes:

[0016] While keeping the weights of other network layers in the multiple network layers unchanged, adjust the weight of the Kth network layer in the multiple network layers according to any second sample data, where K is an integer not less than 2.

[0017] In another possible implementation, the method further includes:

[0018] Under the condition that the weights of other network layers in the multiple network layers remain unchanged, the weight of the K-1th network layer in the multiple network layers is adjusted according to the next second sample data of any one second sample data.

[0019] In another possible implementation, the step of training the prediction model based on at least one first sample data includes:

[0020] For any piece of first sample data, calling the prediction model, and determining the predicted partition coefficient of the first sample compound according to the structural data of the first sample compound in the first sample data;

[0021] The prediction model is trained according to the prediction distribution coefficient and the sample distribution coefficient in the first sample data.

[0022] In another possible implementation, the training of the prediction model according to the prediction allocation coefficient and the sample allocation coefficient in the first sample data includes:

[0023] determining a second loss value of the prediction model according to the prediction allocation coefficient and the sample allocation coefficient in the first sample data;

[0024] The prediction model is trained according to the second loss value.

[0025] In another possible implementation, before training the prediction model based on the at least one first sample data, the method further includes:

[0026] Simulating a motion trajectory of any first sample compound in a mixed solution, wherein the mixed solution is a solution obtained by mixing the first sample compound with a polar solvent and a non-polar solvent;

[0027] The partition coefficient of the first sample compound when thermodynamic equilibrium is reached between the lipid phase and the aqueous phase of the mixed solution is determined as the sample partition coefficient of the first sample compound, wherein the lipid phase is formed by the non-polar solvent and the first sample compound dissolved in the non-polar solvent, and the aqueous phase is formed by the polar solvent and the first sample compound dissolved in the polar solvent.

[0028] In another possible implementation, before training the prediction model based on the at least one first sample data, the method further includes:

[0029] Obtaining a structure string of any first sample compound, the structure string conforming to the Simplified Molecular Linear Input Specification (SMILES), and determining the structure string as the structure data of the first sample compound; or,

[0030] A structural diagram of any first sample compound is obtained, and the structural diagram is determined as structural data of the first sample compound.

[0031] In another possible implementation, the target property parameter of the compound includes the compound's apparent permeability coefficient, solubility in a solvent of the target property, or plasma protein binding rate.

[0032] In another aspect, a device for determining a compound property parameter is provided, the device comprising:

[0033] a first training module, configured to train a prediction model based on at least one first sample data, wherein the first sample data includes structural data and a sample partition coefficient of a first sample compound, wherein the sample partition coefficient is the partition coefficient of the first sample compound between an aqueous phase and a lipid phase, and the trained prediction model is configured to determine the partition coefficient of any compound based on the structural data of the compound;

[0034] a second training module, configured to retrain the trained prediction model based on at least one second sample data, wherein the second sample data includes structural data and sample attribute parameters of a second sample compound, wherein the sample attribute parameters are attribute parameters related to the partition coefficient;

[0035] The property parameter determination module is used to call the retrained prediction model and determine the target property parameter of any compound according to the structural data of any compound.

[0036] In a possible implementation, the second training module includes:

[0037] an attribute parameter prediction unit, configured to call the trained prediction model for any second sample data and determine the predicted attribute parameter of the second sample compound according to the structural data of the second sample compound in the second sample data;

[0038] The first training unit is used to retrain the trained prediction model according to the prediction attribute parameters and the sample attribute parameters in the second sample data.

[0039] In another possible implementation, the first training unit is configured to determine a first loss value of the trained prediction model based on the prediction attribute parameters and the sample attribute parameters in the second sample data; and retrain the trained prediction model based on the first loss value.

[0040] In another possible implementation, the second training module is used to adjust the weight of the Kth network layer among the multiple network layers according to any second sample data while keeping the weights of other network layers among the multiple network layers unchanged, where K is an integer not less than 2.

[0041] In another possible implementation, the second training module is further used to adjust the weight of the K-1th network layer among the multiple network layers according to the next second sample data of any second sample data while keeping the weights of other network layers among the multiple network layers unchanged.

[0042] In another possible implementation, the first training module includes:

[0043] a partition coefficient prediction unit, configured to call the prediction model for any piece of first sample data and determine the predicted partition coefficient of the first sample compound according to the structural data of the first sample compound in the first sample data;

[0044] The second training unit is used to train the prediction model according to the prediction distribution coefficient and the sample distribution coefficient in the first sample data.

[0045] In another possible implementation, the second training unit is configured to determine a second loss value of the prediction model based on the prediction allocation coefficient and the sample allocation coefficient in the first sample data; and train the prediction model based on the second loss value.

[0046] In another possible implementation, the apparatus further includes:

[0047] A partition coefficient determination module is configured to simulate the motion trajectory of any first sample compound in a mixed solution, wherein the mixed solution is a solution obtained by mixing the first sample compound with a polar solvent and a non-polar solvent; and to determine the partition coefficient of the first sample compound when thermodynamic equilibrium is reached between a lipid phase and an aqueous phase of the mixed solution as the sample partition coefficient of the first sample compound, wherein the lipid phase is formed by the non-polar solvent and the first sample compound dissolved in the non-polar solvent, and the aqueous phase is formed by the polar solvent and the first sample compound dissolved in the polar solvent.

[0048] In another possible implementation, the apparatus further includes:

[0049] The structure data acquisition module is used to obtain a structure string of any first sample compound, where the structure string conforms to the Simplified Molecular Linear Input Specification (SMILES), and determine the structure string as the structure data of the first sample compound; or obtain a structure diagram of any first sample compound, and determine the structure diagram as the structure data of the first sample compound.

[0050] In another possible implementation, the target property parameter of the compound includes the compound's apparent permeability coefficient, solubility in a solvent of the target property, or plasma protein binding rate.

[0051] On the other hand, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one computer program, and the computer program is loaded and executed by the processor to implement the operations performed in the compound property parameter determination method in any of the above possible implementations.

[0052] On the other hand, a computer-readable storage medium is provided, in which at least one computer program is stored. The computer program is loaded and executed by a processor to implement the operations performed in the compound property parameter determination method in any of the possible implementations described above.

[0053] In another aspect, a computer program product or computer program is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the operations performed in the compound property parameter determination method of the various optional implementations described above.

[0054] The beneficial effects of the technical solutions provided in the embodiments of the present application include at least:

[0055] In the technical solution provided in the embodiments of the present application, for the attribute parameters related to the partition coefficient of the compound, the prediction model is not trained directly using the attribute parameters of the sample compound. Instead, the prediction model is first trained based on the partition coefficient of the sample compound, and then retrained using the attribute parameters of the compound based on the trained prediction model. Since the attribute parameter is related to the partition coefficient of the compound, for the prediction model, predicting the partition coefficient and predicting the attribute parameter are similar tasks. Therefore, under the premise that the prediction model has the ability to predict the partition coefficient, it is only necessary to retrain the prediction model using the attribute parameters of a small number of compounds. The model can have a higher attribute parameter prediction ability without obtaining the attribute parameters of a large number of compounds. This allows the attribute parameters of the compound to be determined based on the retrained prediction model, thereby improving the accuracy of determining the attribute parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0057] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present application;

[0058] Figure 2 This is a schematic diagram of a drug development process provided in an embodiment of the present application;

[0059] Figure 3 This is a flow chart of a method for determining compound property parameters provided in an embodiment of the present application;

[0060] Figure 4 This is a schematic diagram of the relationship between solubility and partition coefficient provided in an embodiment of the present application;

[0061] Figure 5This is a flow chart of a method for determining compound property parameters provided in an embodiment of the present application;

[0062] Figure 6 This is a schematic diagram of a training process of a prediction model provided in an embodiment of the present application;

[0063] Figure 7 This is a schematic diagram of a model prediction effect provided in an embodiment of the present application;

[0064] Figure 8 This is a schematic diagram of a model prediction effect provided in an embodiment of the present application;

[0065] Figure 9 This is a block diagram of a device for determining compound property parameters provided by an embodiment of the present application;

[0066] Figure 10 This is a block diagram of a device for determining compound property parameters provided by an embodiment of the present application;

[0067] Figure 11 This is a schematic diagram of the structure of a terminal provided in an embodiment of the present application;

[0068] Figure 12 This is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0070] As used herein, the terms "first," "second," "third," and "fourth" may be used to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are used solely to distinguish one concept from another. For example, without departing from the scope of this application, first sample data may be referred to as sample data, and similarly, second sample data may be referred to as first sample data.

[0071] As used herein, the terms "at least one," "plurality," "each," and "any" include one, two, or more, "plurality" includes two or more, "each" refers to each of the corresponding plurality, and "any" refers to any one of the plurality. For example, if the plurality of sample compounds includes three sample compounds, "each" refers to each of the three sample compounds, and "any" refers to any one of the three sample compounds, which may be the first, second, or third.

[0072] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present application. Figure 1 The implementation environment includes a terminal 101 and a server 102. Terminal 101 and server 102 are connected via a wireless or wired network. Optionally, terminal 101 is a computer, mobile phone, tablet computer, or other terminal. Optionally, server 102 is a backend server for the target application or a cloud server that provides cloud computing and cloud storage services.

[0073] Optionally, a target application provided by the server 102 is installed on the terminal 101, and the terminal 101 can implement functions such as data transmission and message interaction through the target application. Optionally, the target application is a target application in the terminal 101 operating system, or a target application provided by a third party. The target application has the function of determining the target attribute parameters of the compound, for example, the target application can determine the solubility, apparent permeability coefficient, plasma protein binding rate, etc. of the compound. Optionally, the target application also has the function of determining other attribute parameters of the compound, which is not limited by this application.

[0074] In the embodiment of the present application, the server 102 is used to train a prediction model and send the trained prediction model to the terminal 101, and the terminal 101 is used to call the prediction model to determine the target attribute parameters of any compound.

[0075] The compound attribute parameter determination method provided in this application can be applied to any scenario for determining the target attribute parameters of a compound. For example, it can be applied to the scenario of drug development. When conducting drug development, for a variety of candidate compounds for preparing drugs, the method provided in this application can be used to determine the target attribute parameters of each candidate compound, such as solubility, apparent permeability coefficient, plasma protein binding rate, etc. Since these three attribute parameters determine the absorption effect of the drug after the candidate compound is prepared into a drug, it is possible to determine the compound ultimately used for pharmaceutical production, i.e., the lead compound, based on these three attribute parameters of each candidate compound.

[0076] Figure 2 This is a schematic diagram of the drug development process. Figure 2, first target identification and verification are carried out, that is, the binding site of the candidate compound and the body's biological macromolecules is identified, and the binding site is verified. Then, the candidate compound is screened to determine the lead compound ultimately used for pharmaceutical production. Preclinical research and clinical trials are then carried out. Among them, when the candidate compound is screened, the properties such as absorption (Absorption), distribution (Distribution), metabolism (Metabolism), excretion (Excretion) and toxicity (Toxicity) of the candidate compound are determined, so that R&D personnel can be helped to screen out the lead compound ultimately used for pharmaceutical production and improve R&D efficiency. And in the process of research and development, the property parameters determined by the compound property parameter determination method provided in this application can help R&D personnel predict the absorption properties after the compound is made into medicine.

[0077] Another example is the scenario applied to environmental chemistry research. For some compounds with polluting properties, their diffusion in the environment will cause harm to the environment and various organisms, and the solubility of the compound determines its diffusion ability in the environment and in the organism. Therefore, the solubility of various compounds can be determined by the compound property parameter determination method provided in the embodiment of the present application, and then the harmfulness of the compound to the environment can be evaluated based on the solubility of the compound. Alternatively, in the case where a certain compound with polluting properties leaks into the environment, the solubility of the compound can be determined by the method provided in the present application, and then the pollution status of the compound can be predicted based on the solubility.

[0078] Figure 3 This is a flow chart of a method for determining compound property parameters provided in an embodiment of the present application. The execution subject of this embodiment is a computer device. Figure 3 , the embodiment includes:

[0079] 301. Train a prediction model based on at least one first sample data, where the first sample data includes structural data and a sample partition coefficient of a first sample compound, where the sample partition coefficient is the partition coefficient of the first sample compound between an aqueous phase and a lipid phase. The trained prediction model is used to determine the partition coefficient of any compound based on the structural data of the compound.

[0080] A compound is a compound composed of molecules, including neutral, acidic, and basic compounds. The structural data of a compound is used to represent the compound's components and the structural relationships between them. The partition coefficient is the partition coefficient of a neutral compound between an aqueous phase and a lipid phase at a pH (potential of hydrogen) of 7.

[0081] It is important to note that after mixing a compound with a polar solvent and a non-polar solvent, the resulting mixed solution will separate into two layers, since polar solvents and non-polar solvents are immiscible. These layers are the lipid phase, formed by the non-polar solvent and the compound dissolved in the non-polar solvent, and the aqueous phase, formed by the polar solvent and the compound dissolved in the polar solvent. Because different compounds have different solubility properties—for example, some are hydrophilic, meaning they dissolve easily in polar solvents, while others are lipophilic, meaning they dissolve easily in non-polar solvents—when the compound reaches thermodynamic equilibrium between the lipid and aqueous phases—that is, when the compound's concentrations in the aqueous and lipid phases no longer change over time—the concentrations of the compound in the aqueous and lipid phases may differ. The ratio of the compound's concentration in the lipid phase to its concentration in the aqueous phase is called the lipid-water partition coefficient. A larger lipid-water partition coefficient indicates that the compound is more soluble in non-polar solvents, while a smaller lipid-water partition coefficient indicates that the compound is more soluble in polar solvents. In this application, the partition coefficient includes the fat-water partition coefficient and the logarithm of the fat-water partition coefficient. For example, if P represents the fat-water partition coefficient, the partition coefficient in this application includes P and LogP.

[0082] 302. Retrain the trained prediction model according to at least one second sample data, where the second sample data includes structural data and sample attribute parameters of the second sample compound, where the sample attribute parameters are attribute parameters related to the partition coefficient.

[0083] Compounds have various types of attribute parameters. For example, the above-mentioned distribution coefficient is one type of attribute parameter. In addition, compounds also have other types of attribute parameters, such as apparent permeability coefficient, plasma protein binding rate, etc. The sample attribute parameter is any attribute parameter related to the distribution coefficient among the various types of attribute parameters of the compound. For example, the sample attribute parameter is the solubility, apparent permeability coefficient or plasma protein binding rate of the compound in the solvent of the target attribute. Among them, the solvent of the target attribute includes polar solvents and non-polar solvents. It should be noted that the sample attribute parameter of the second sample compound can also be the distribution coefficient of the second sample compound under pH conditions other than pH 7. For example, it is the distribution coefficient of the second sample compound under pH 7.4.

[0084] The apparent permeability coefficient is a parameter that describes the membrane permeability of a compound. The larger its value, the stronger the ability of the compound to penetrate the cell membrane. The plasma protein binding rate refers to the percentage of the compound that binds to the plasma proteins in the blood after being absorbed into the blood. Solubility indicates the maximum amount of a substance that dissolves in a certain amount of solvent under a certain temperature and pressure. It is generally expressed in grams of a substance that can be dissolved in 100 grams of solvent. Solubility includes thermodynamic solubility and kinetic solubility. In this application, if the sample attribute parameter of the second sample compound in the second sample data is thermodynamic solubility, the retrained prediction model is used to predict the thermodynamic solubility of any compound. If the sample attribute parameter of the second sample compound in the second sample data is kinetic solubility, the retrained prediction model is used to predict the kinetic solubility of any compound.

[0085] Figure 4 Schematic diagram of the relationship between the solubility and distribution coefficient of a compound, refer to Figure 4 Each point in the coordinate graph represents a compound. The horizontal coordinate of the point is the negative distribution coefficient LogP of the compound, i.e. -LogP. The vertical coordinate of the point is the logarithm of the solubility of the compound in polar solution. S represents the solubility of the compound in polar solution, and the vertical coordinate of the point is LogS. From the straight line fitted by multiple points, it can be seen that the LogS of the compound is positively correlated with -LogP, and the Pearson correlation coefficient (Pearson R 2 ) is 0.6972, and the root mean square error (RMSE) is 1.1489, which also shows that the two have a strong correlation. On this basis, it can be determined that the distribution coefficient of the compound has a strong correlation with the solubility.

[0086] The process of training the prediction model in this application is based on the correlation between the partition coefficient of the compound and the attribute parameters. In this application, the prediction model is first trained by the partition coefficient of the compound so that the model has the ability to predict the partition coefficient of the compound. Since the partition coefficient of the compound is easy to obtain, the prediction model can be trained by the partition coefficient of a large number of compounds, so that the accuracy of the trained prediction model in predicting the partition coefficient is high. Since the sample attribute parameters are related to the partition coefficient of the compound, for the prediction model, predicting the partition coefficient of the compound and predicting the sample attribute parameters are similar tasks. Therefore, when the prediction model can accurately predict the partition coefficient, it is only necessary to retrain the prediction model by the sample attribute parameters of a small number of compounds. The prediction model can accurately predict the target attribute parameters of the compound of the same type as the sample attribute parameters. In this way, the problem of insufficient generalization and low accuracy of the trained prediction model due to the small number of attribute parameters of the compound is solved.

[0087] One thing that needs to be explained is that the process of training the prediction model mentioned above actually utilizes transfer learning technology. Transfer learning refers to the process of transferring the weights of the network layer in a neural network from an already trained neural network to a new neural network used to handle different tasks. In other words, the new neural network used to handle different tasks is not trained from scratch, but is trained on the basis of the neural network that has been trained to handle other tasks. The tasks handled by the already trained neural network and the new neural network are similar. Since the tasks of the two neural networks are not very different, many of the underlying neurons handle the same things. Therefore, retraining the already trained neural network to obtain a neural network used to handle new tasks can increase the training speed of the neural network and reduce the amount of data required for training.

[0088] 303. Call the retrained prediction model to determine the target attribute parameters of any compound based on the structural data of the compound.

[0089] The target attribute parameter and the sample attribute parameter of the sample compound are attribute parameters of the same type. For example, if the sample attribute parameter is the solubility of the sample compound in the solvent of the target attribute, then the target attribute parameter is the solubility of any compound in the solvent of the target attribute.

[0090] In the technical solution provided in the embodiments of the present application, for the attribute parameters related to the partition coefficient of the compound, the prediction model is not trained directly using the attribute parameters of the sample compound. Instead, the prediction model is first trained based on the partition coefficient of the sample compound, and then retrained using the attribute parameters of the compound based on the trained prediction model. Since the attribute parameter is related to the partition coefficient of the compound, for the prediction model, predicting the partition coefficient and predicting the attribute parameter are similar tasks. Therefore, under the premise that the prediction model has the ability to predict the partition coefficient, it is only necessary to retrain the prediction model using the attribute parameters of a small number of compounds. The model can have a higher attribute parameter prediction ability without obtaining the attribute parameters of a large number of compounds. This allows the attribute parameters of the compound to be determined based on the retrained prediction model, thereby improving the accuracy of determining the attribute parameters.

[0091] Figure 5 This is a flow chart of a method for determining compound property parameters provided in an embodiment of the present application. Figure 5 , the embodiment includes:

[0092] 501. A terminal obtains at least one piece of first sample data, where the first sample data includes structural data and a sample partition coefficient of a first sample compound. The sample partition coefficient is a partition coefficient of the first sample compound between an aqueous phase and a lipid phase.

[0093] There are two implementations of the terminal acquiring the structural data of the first sample compound.

[0094] First, the terminal obtains a structure string of any first sample compound, the structure string conforming to SMILES (Simplified molecular input line entry specification), and determines the structure string as the structure data of the first sample compound.

[0095] SMILES is a standard that uses ASCII strings to clearly describe molecular structures. Each molecule has only one structure string that conforms to SMILES. Therefore, after the model obtains the structure string that conforms to SMILES for each compound, it can distinguish different compounds based on each compound's structure string.

[0096] The second method is to obtain a structural diagram of any first sample compound and determine the structural diagram as the structural data of the first sample compound. Optionally, the structural diagram is a two-dimensional structural diagram of the first sample compound.

[0097] Optionally, the terminal obtains structural data of multiple compounds from a molecular database. The molecular database stores structural data of multiple compounds, such as structural strings of multiple compounds that conform to SMILES. Optionally, the molecular database includes ChEMBL (a molecular database) or ZINC (a molecular database).

[0098] In one possible implementation, the terminal obtains the sample partition coefficient of the first sample compound, including: the terminal simulates the movement trajectory of any first sample compound in the mixed solution; and determines the partition coefficient of the first sample compound when the first sample compound reaches thermodynamic equilibrium between the lipid phase and the water phase of the mixed solution as the sample partition coefficient of the first sample compound.

[0099] The mixed solution is a solution obtained by mixing a first sample compound with a polar solvent and a non-polar solvent. The lipid phase is formed by the non-polar solvent and the first sample compound dissolved in the non-polar solvent, and the aqueous phase is formed by the polar solvent and the first sample compound dissolved in the polar solvent. After the first sample compound is mixed with the polar solvent and the non-polar solvent, the compound diffuses in the mixed solution. The present application directly simulates the motion trajectory of the first sample compound in the mixed solution through a terminal. Based on this motion trajectory, the diffusion position of the compound in the mixed solution at any time can be determined, thereby determining the concentration of the first sample compound in the lipid phase and the aqueous phase of the mixed solution at any time. When the first sample compound reaches thermodynamic equilibrium between the lipid phase and the aqueous phase, that is, when the concentrations of the first sample compound in both the lipid phase and the aqueous phase no longer change, the ratio of the concentration of the first sample compound in the lipid phase to the concentration in the aqueous phase is determined as the sample partition coefficient of the first sample compound.

[0100] In the embodiments of the present application, the distribution coefficient of the compound can be determined quickly and conveniently by simulating the movement trajectory of the compound in the mixed solution. When the number of sample compounds is large, the efficiency of determining the distribution coefficient is greatly improved.

[0101] It should be noted that the above-mentioned method for determining the distribution coefficient of the compound actually adopts the molecular dynamics method. Of course, other methods can also be used, such as quantum mechanics method, fragment addition method, Wildman-Crippen (a method for determining the distribution coefficient of a compound), etc. to determine the distribution coefficient of the compound. The embodiments of this application are not limited to this.

[0102] Another point worth noting is that, optionally, the terminal uses each compound in the molecular database as a sample compound, determines the partition coefficient for each sample compound, and then generates first sample data based on the structural data and partition coefficient for each sample compound. Because the number of compounds in the molecular database is very large, the terminal can obtain a large amount of first sample data, making the trained prediction model more accurate.

[0103] 502. The terminal trains a prediction model according to at least one first sample data.

[0104] The prediction model trained according to the first sample data is used to determine the partition coefficient of any compound according to the structural data of the compound.

[0105] In one possible implementation, the terminal trains a prediction model based on at least one first sample data, including: for any first sample data, the terminal calls the prediction model, determines the predicted distribution coefficient of the first sample compound based on the structural data of the first sample compound in the first sample data; and trains the prediction model based on the predicted distribution coefficient and the sample distribution coefficient in the first sample data.

[0106] The terminal trains the prediction model based on the predicted allocation coefficient and the sample allocation coefficient in the first sample data so that the predicted allocation coefficient and the sample allocation coefficient are continuously approximated until a reference condition is satisfied. For example, the terminal adjusts weights of a network layer in the prediction model so that the predicted allocation coefficient and the sample allocation coefficient are continuously approximated until the reference condition is satisfied.

[0107] In one possible implementation, the terminal trains a prediction model based on the prediction allocation coefficient and the sample allocation coefficient in the first sample data, including: the terminal determines a second loss value of the prediction model based on the prediction allocation coefficient and the sample allocation coefficient in the first sample data; and trains the prediction model based on the second loss value.

[0108] The second loss value can represent the prediction accuracy of the prediction model. The smaller the second loss value, the higher the prediction accuracy of the prediction model and the more accurate the partition coefficient of the determined compound. Optionally, the second loss value is the absolute value of the difference between the predicted partition coefficient and the sample partition coefficient.

[0109] The terminal trains the prediction model based on the second loss value so that the second loss value corresponding to the prediction allocation coefficient determined based on the trained prediction model continuously decreases until the second loss value is less than the reference loss value. In this way, the prediction accuracy of the trained prediction model can be guaranteed.

[0110] 503. The terminal obtains at least one piece of second sample data, where the second sample data includes structural data and sample attribute parameters of the second sample compound, where the sample attribute parameters are attribute parameters related to the partition coefficient.

[0111] It should be noted that the second sample compound in the second sample data may be the same as or different from the first sample compound in the first sample data, and this embodiment of the present application does not impose any limitation on this.

[0112] 504. The terminal retrains the trained prediction model according to at least one second sample data.

[0113] In one possible implementation, the terminal retrains the trained prediction model based on at least one second sample data, including: for any second sample data, the terminal calls the trained prediction model, determines the predicted attribute parameters of the second sample compound based on the structural data of the second sample compound in the second sample data; and retrains the trained prediction model based on the predicted attribute parameters and the sample attribute parameters in the second sample data.

[0114] The terminal retrains the trained prediction model based on the predicted attribute parameters and the sample attribute parameters in the second sample data, so that the predicted attribute parameters are continuously close to the sample attribute parameters until the reference condition is satisfied. For example, the terminal adjusts the weights of the network layers in the trained prediction model so that the predicted attribute parameters are continuously close to the sample attribute parameters until the reference condition is satisfied.

[0115] In one possible implementation, the terminal retrains the trained prediction model based on the prediction attribute parameters and the sample attribute parameters in the second sample data, including: the terminal determines a first loss value of the trained prediction model based on the prediction attribute parameters and the sample attribute parameters in the second sample data; and retrains the trained prediction model based on the first loss value.

[0116] The first loss value can represent the prediction accuracy of the prediction model. The smaller the first loss value, the higher the prediction accuracy of the prediction model and the more accurate the determined property parameter of the compound. Optionally, the first loss value is the absolute value of the difference between the predicted property parameter and the sample property parameter.

[0117] The terminal retrains the trained prediction model according to the first loss value, so that the first loss value corresponding to the prediction attribute parameter determined according to the retrained prediction model continues to decrease until the first loss value is less than the reference loss value.

[0118] In one possible implementation, the prediction model trained based on the first sample data includes multiple network layers. Accordingly, the terminal retrains the trained prediction model based on at least one piece of second sample data, including: adjusting the weight of a Kth network layer among the multiple network layers based on any piece of second sample data while maintaining the weights of other network layers among the multiple network layers unchanged, where K is an integer not less than 2.

[0119] Optionally, K is equal to the number of network layers, and the Kth network layer is the last layer of the multiple network layers, for example, the output layer. Since predicting the property parameters of a compound based on its structural data and predicting the distribution coefficient of a compound based on its structural data are similar tasks, the features extracted by the network layer close to the input layer of the prediction model used to determine the property parameters may be the same as those of the prediction model used to determine the distribution coefficient. This means that the weights of the first few network layers of the prediction model used to determine the property parameters and the prediction model used to determine the distribution coefficient are the same, and only the weights of the output layer are different. Therefore, in the above method, the terminal only adjusts the weight of the last network layer while keeping the weights of the other network layers unchanged, thereby improving the efficiency of model training while ensuring the training effect.

[0120] Optionally, the terminal adjusts the weight of the Kth network layer among the multiple network layers based on any second sample data by: the terminal calls a trained prediction model, determines a predicted attribute parameter of the second sample compound based on the structural data of the second sample compound in the second sample data, and adjusts the weight of the Kth network layer among the multiple network layers based on the predicted attribute parameter and the sample attribute parameter in the second sample data. Optionally, the terminal adjusts the weight of the Kth network layer among the multiple network layers based on the predicted attribute parameter and the sample attribute parameter in the second sample data by: the terminal adjusts the weight of the Kth network layer until the similarity between the predicted attribute parameter determined according to the adjusted prediction model and the sample attribute parameter reaches a reference condition.

[0121] In one possible implementation, the prediction model used to determine the attribute parameters may have different weights for multiple network layers close to the output layer than the prediction model used to determine the allocation coefficient. Therefore, after the terminal adjusts the weight of the Kth network layer according to any second sample data, while keeping the weights of other network layers in the multiple network layers unchanged, it adjusts the weight of the K-1th network layer in the multiple network layers according to the next second sample data of any second sample data. In this way, the terminal adjusts the weights of the multiple network layers layer by layer according to the multiple second sample data in order from back to front, thereby improving the efficiency of model training while ensuring the training effect. Among them, the implementation method of the terminal adjusting the weight of the K-1th network layer is the same as that of adjusting the weight of the Kth network layer, which will not be repeated here.

[0122] It should be noted that training the prediction model based on the first sample data can be referred to as pre-training the prediction model, and the trained prediction model can be referred to as the pre-trained model. Retraining the trained prediction model based on the second sample data can be referred to as fine-tuning the pre-trained model. The amount of the second sample data is smaller than the amount of the first sample data.

[0123] Figure 6 This is a diagram of the training process of the prediction model. Figure 6 First, the partition coefficients of the compounds in the molecular database are determined. A prediction model is then trained based on the structural data and partition coefficients of a large number of sample compounds in the molecular database. The task of the trained prediction model, or the original task, is to determine the partition coefficient of any compound based on its structural data. The trained prediction model is then retrained based on the structural data and sample attribute parameters of a small number of sample compounds, i.e., the weights of the trained prediction model are adjusted. The task of the retrained prediction model, or the target task, is to determine the target attribute parameters of any compound based on its structural data. These target attribute parameters belong to the same type of attribute parameters as the sample attribute parameters.

[0124] 505. The terminal calls the retrained prediction model and determines the target attribute parameters of any compound according to the structural data of any compound.

[0125] After retraining the trained prediction model based on the second sample data, the retrained prediction model can determine the target attribute parameter of any compound based on the structural data of the compound. The target attribute parameter and the sample attribute parameter of the second sample compound are of the same type of attribute parameter. For example, both are the solubility of the compound or the apparent permeability coefficient of the compound.

[0126] Optionally, after retraining the trained prediction model, the terminal uploads the retrained prediction model to the server, determines a call interface for the prediction model, and calls the retrained prediction model through the call interface when determining the target property parameters of any compound. Because the prediction model is relatively large and occupies a large amount of storage space, this method eliminates the need for the terminal to store the prediction model, thereby saving terminal resources.

[0127] It should be noted that the execution entity of the embodiment of the present application can be other computer devices besides the terminal, for example, a server, and the embodiment of the present application does not limit this.

[0128] Figure 7 and Figure 8 This is a comparison chart of the prediction effects of the prediction model in this application and the prediction model in the related art. Among them, the prediction model in the related art uses randomly generated weights as the starting point for prediction model training, and then is trained using the property parameters of the compound. Figure 7 In the Pearson correlation coefficient, the correlation between the property parameters determined by the prediction model and the actual property parameters of the compound is expressed. The larger the Pearson correlation coefficient, the higher the accuracy of the prediction model. Figure 7 It can be seen that the prediction model in this application is better than the prediction model in the related art in predicting multiple attribute parameters such as the kinetic solubility, thermodynamic solubility, the apparent permeability coefficient of the compound for MDCK (Madin–Darby canine kidney, canine kidney epithelial cells), the BRPP (binding rate of plasma protein, plasma protein binding rate) of the compound, and the apparent permeability coefficient of the compound for Caco-2 (human intestinal epithelial cells). Figure 8 The root mean square error in represents the root mean square error between the property parameters determined by the prediction model and the actual property parameters of the compound. The smaller the root mean square error, the higher the accuracy of the prediction model. Figure 8 It can be seen that the prediction model in this application is better than the prediction model in the related art in predicting multiple property parameters such as the kinetic solubility, thermodynamic solubility, apparent permeability coefficient of the compound for MDCK, BRPP of the compound, apparent permeability coefficient of the compound for Caco-2, and partition coefficient of the compound at pH 7.4.

[0129] In the technical solution provided in the embodiments of the present application, for the attribute parameters related to the partition coefficient of the compound, the prediction model is not trained directly using the attribute parameters of the sample compound. Instead, the prediction model is first trained based on the partition coefficient of the sample compound, and then retrained using the attribute parameters of the compound based on the trained prediction model. Since the attribute parameter is related to the partition coefficient of the compound, for the prediction model, predicting the partition coefficient and predicting the attribute parameter are similar tasks. Therefore, under the premise that the prediction model has the ability to predict the partition coefficient, it is only necessary to retrain the prediction model using the attribute parameters of a small number of compounds. The model can have a higher attribute parameter prediction ability without obtaining the attribute parameters of a large number of compounds. This allows the attribute parameters of the compound to be determined based on the retrained prediction model, thereby improving the accuracy of determining the attribute parameters.

[0130] All of the above optional technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0131] Figure 9 This is a block diagram of a device for determining compound property parameters provided by an embodiment of the present application. Figure 9 , the device comprises:

[0132] A first training module 901 is configured to train a prediction model based on at least one first sample data, wherein the first sample data includes structural data and a sample partition coefficient of a first sample compound, wherein the sample partition coefficient is the partition coefficient of the first sample compound between an aqueous phase and a lipid phase. The trained prediction model is configured to determine the partition coefficient of any compound based on the structural data of any compound.

[0133] A second training module 902 is configured to retrain the trained prediction model based on at least one second sample data, wherein the second sample data includes structural data and sample attribute parameters of the second sample compound, wherein the sample attribute parameters are attribute parameters related to the partition coefficient;

[0134] The attribute parameter determination module 903 is used to call the retrained prediction model and determine the target attribute parameter of any compound according to the structural data of any compound.

[0135] In one possible implementation, refer to Figure 10 The second training module 902 includes:

[0136] The property parameter prediction unit 9021 is used to call the trained prediction model for any second sample data and determine the predicted property parameter of the second sample compound according to the structural data of the second sample compound in the second sample data;

[0137] The first training unit 9022 is used to retrain the trained prediction model according to the prediction attribute parameters and the sample attribute parameters in the second sample data.

[0138] In another possible implementation, the first training unit 9022 is configured to determine a first loss value of the trained prediction model based on the prediction attribute parameters and the sample attribute parameters in the second sample data; and retrain the trained prediction model based on the first loss value.

[0139] In another possible implementation, the second training module 902 is used to adjust the weight of the Kth network layer among the multiple network layers according to any second sample data while keeping the weights of other network layers among the multiple network layers unchanged, where K is an integer not less than 2.

[0140] In another possible implementation, the second training module 902 is further used to adjust the weight of the K-1th network layer among the multiple network layers according to the next second sample data of any second sample data while keeping the weights of other network layers among the multiple network layers unchanged.

[0141] In another possible implementation, continue to refer to Figure 10 The first training module 901 includes:

[0142] The partition coefficient prediction unit 9011 is configured to call the prediction model for any piece of first sample data and determine the predicted partition coefficient of the first sample compound according to the structural data of the first sample compound in the first sample data;

[0143] The second training unit 9012 is used to train the prediction model according to the prediction distribution coefficient and the sample distribution coefficient in the first sample data.

[0144] In another possible implementation, the second training unit 9012 is configured to determine a second loss value of the prediction model based on the prediction allocation coefficient and the sample allocation coefficient in the first sample data; and train the prediction model based on the second loss value.

[0145] In another possible implementation, continue to refer to Figure 10 , the device further comprises:

[0146] The partition coefficient determination module 904 is configured to simulate the motion trajectory of any first sample compound in a mixed solution, where the mixed solution is a mixture of the first sample compound, a polar solvent, and a non-polar solvent; and to determine the partition coefficient of the first sample compound when thermodynamic equilibrium is reached between the lipid phase and the aqueous phase of the mixed solution as the sample partition coefficient of the first sample compound, where the lipid phase is formed by the non-polar solvent and the first sample compound dissolved in the non-polar solvent, and the aqueous phase is formed by the polar solvent and the first sample compound dissolved in the polar solvent.

[0147] In another possible implementation, continue to refer to Figure 10 , the device further comprises:

[0148] The structure data acquisition module 905 is used to obtain a structure string of any first sample compound, where the structure string conforms to the Simplified Molecular Linear Input Standard (SMILES), and determine the structure string as the structure data of the first sample compound; or obtain a structure diagram of any first sample compound and determine the structure diagram as the structure data of the first sample compound.

[0149] In another possible implementation, the target property parameter of the compound includes the compound's apparent permeability coefficient, solubility in a solvent of the target property, or plasma protein binding rate.

[0150] In the technical solution provided in the embodiments of the present application, for the attribute parameters related to the partition coefficient of the compound, the prediction model is not trained directly using the attribute parameters of the sample compound. Instead, the prediction model is first trained based on the partition coefficient of the sample compound, and then retrained using the attribute parameters of the compound based on the trained prediction model. Since the attribute parameter is related to the partition coefficient of the compound, for the prediction model, predicting the partition coefficient and predicting the attribute parameter are similar tasks. Therefore, under the premise that the prediction model has the ability to predict the partition coefficient, it is only necessary to retrain the prediction model using the attribute parameters of a small number of compounds. The model can have a higher attribute parameter prediction ability without obtaining the attribute parameters of a large number of compounds. This allows the attribute parameters of the compound to be determined based on the retrained prediction model, thereby improving the accuracy of determining the attribute parameters.

[0151] It should be noted that the above-described embodiments of the compound property parameter determination apparatus, when determining the solubility of a compound, illustrate only the division of the aforementioned functional modules. In actual applications, the aforementioned functions can be assigned to different functional modules as needed, i.e., the internal structure of a computer device can be divided into different functional modules to perform all or part of the functions described above. Furthermore, the compound property parameter determination apparatus provided in the above-described embodiments and the compound property parameter determination method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0152] An embodiment of the present application also provides a computer device, which includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed in the compound property parameter determination method of the above embodiment.

[0153] Optionally, the computer device is provided as a terminal. Figure 11The following is a block diagram of a terminal 1100 according to an exemplary embodiment of the present application. Terminal 1100 may be a smartphone, tablet computer, MP3 player (Moving Picture Experts Group Audio Layer III), MP4 player (Moving Picture Experts Group Audio Layer IV), laptop computer, or desktop computer. Terminal 1100 may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other similar names.

[0154] The terminal 1100 includes a processor 1101 and a memory 1102 .

[0155] The processor 1101 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1101 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 1101 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1101 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1101 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0156] Memory 1102 may include one or more computer-readable storage media, which may be non-transitory. Memory 1102 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in memory 1102 is used to store at least one computer program, which is executed by processor 1101 to implement the compound property parameter determination method provided in the method embodiment of the present application.

[0157] In some embodiments, terminal 1100 may optionally include a peripheral device interface 1103 and at least one peripheral device. Processor 1101, memory 1102, and peripheral device interface 1103 may be connected via a bus or signal lines. Each peripheral device may be connected to peripheral device interface 1103 via a bus, signal lines, or circuit boards. Specifically, the peripheral device may include at least one of a radio frequency circuit 1104, a display screen 1105, a camera assembly 1106, an audio circuit 1107, a positioning assembly 1108, and a power supply 1109.

[0158] The peripheral device interface 1103 can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor 1101 and the memory 1102. In some embodiments, the processor 1101, the memory 1102, and the peripheral device interface 1103 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1101, the memory 1102, and the peripheral device interface 1103 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0159] The RF circuit 1104 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 1104 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 1104 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the RF circuit 1104 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The RF circuit 1104 can communicate with other terminals via at least one wireless communication protocol. Such wireless communication protocols include, but are not limited to, metropolitan area networks, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 1104 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.

[0160] The display screen 1105 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1105 is a touch screen display, the display screen 1105 also has the ability to collect touch signals on the surface or above the surface of the display screen 1105. The touch signal can be input as a control signal to the processor 1101 for processing. At this time, the display screen 1105 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, there can be one display screen 1105, which is set on the front panel of the terminal 1100; in other embodiments, there can be at least two display screens 1105, which are respectively set on different surfaces of the terminal 1100 or in a folding design; in other embodiments, the display screen 1105 can be a flexible display screen, which is set on the curved surface or folding surface of the terminal 1100. Even more, the display screen 1105 can be set as a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 1105 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0161] The camera assembly 1106 is used to capture images or videos. Optionally, the camera assembly 1106 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1106 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0162] The audio circuit 1107 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input into the processor 1101 for processing, or input into the radio frequency circuit 1104 to achieve voice communication. For the purpose of stereo sound collection or noise reduction, there may be multiple microphones, each located in different parts of the terminal 1100. The microphone may also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert electrical signals from the processor 1101 or the radio frequency circuit 1104 into sound waves. The speaker may be a traditional thin film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert electrical signals into sound waves audible to humans, but also convert electrical signals into sound waves inaudible to humans for purposes such as distance measurement. In some embodiments, the audio circuit 1107 may also include a headphone jack.

[0163] Positioning component 1108 is used to locate the current geographic location of terminal 1100 to implement navigation or LBS (Location Based Service). Positioning component 1108 can be based on the US GPS (Global Positioning System), China's BeiDou system, Russia's Greninja system, or the European Union's Galileo system.

[0164] Power supply 1109 is used to power various components in terminal 1100. Power supply 1109 can be AC power, DC power, a disposable battery, or a rechargeable battery. When power supply 1109 includes a rechargeable battery, the rechargeable battery can support wired charging or wireless charging. The rechargeable battery can also be used to support fast charging technology.

[0165] In some embodiments, the terminal 1100 further includes one or more sensors 1110 , including but not limited to: an acceleration sensor 1111 , a gyroscope sensor 1112 , a pressure sensor 1113 , a fingerprint sensor 1114 , an optical sensor 1115 , and a proximity sensor 1116 .

[0166] The accelerometer 1111 can detect the magnitude of acceleration along the three coordinate axes of the coordinate system established by the terminal 1100. For example, the accelerometer 1111 can be used to detect the components of gravity acceleration along the three coordinate axes. The processor 1101 can control the display screen 1105 to display the user interface in a landscape or portrait view based on the gravity acceleration signal collected by the accelerometer 1111. The accelerometer 1111 can also be used to collect game or user motion data.

[0167] The gyroscope sensor 1112 can detect the orientation and rotation angle of the terminal 1100. The gyroscope sensor 1112 can work with the acceleration sensor 1111 to collect the user's 3D movements on the terminal 1100. Based on the data collected by the gyroscope sensor 1112, the processor 1101 can implement the following functions: motion sensing (such as changing the UI based on the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0168] The pressure sensor 1113 can be set on the side frame of the terminal 1100 and / or the lower layer of the display screen 1105. When the pressure sensor 1113 is set on the side frame of the terminal 1100, it can detect the user's grip signal of the terminal 1100, and the processor 1101 performs left and right hand recognition or shortcut operations based on the grip signal collected by the pressure sensor 1113. When the pressure sensor 1113 is set on the lower layer of the display screen 1105, the processor 1101 controls the operable controls on the UI interface based on the user's pressure operation on the display screen 1105. The operable controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0169] The fingerprint sensor 1114 is used to collect the user's fingerprint. The processor 1101 identifies the user's identity based on the fingerprint collected by the fingerprint sensor 1114, or the fingerprint sensor 1114 identifies the user's identity based on the collected fingerprint. When the user's identity is recognized as a trusted identity, the processor 1101 authorizes the user to perform relevant sensitive operations, such as unlocking the screen, viewing encrypted information, downloading software, making payments, and changing settings. The fingerprint sensor 1114 can be set on the front, back, or side of the terminal 1100. When a physical button or manufacturer logo is provided on the terminal 1100, the fingerprint sensor 1114 can be integrated with the physical button or manufacturer logo.

[0170] Optical sensor 1115 is used to collect ambient light intensity. In one embodiment, processor 1101 can control the display brightness of display screen 1105 based on the ambient light intensity collected by optical sensor 1115. Specifically, when the ambient light intensity is high, the display brightness of display screen 1105 is increased; when the ambient light intensity is low, the display brightness of display screen 1105 is decreased. In another embodiment, processor 1101 can also dynamically adjust the shooting parameters of camera assembly 1106 based on the ambient light intensity collected by optical sensor 1115.

[0171] Proximity sensor 1116, also known as a distance sensor, is disposed on the front panel of terminal 1100. Proximity sensor 1116 is used to detect the distance between the user and the front of terminal 1100. In one embodiment, when proximity sensor 1116 detects that the distance between the user and the front of terminal 1100 is gradually decreasing, processor 1101 controls display screen 1105 to switch from the screen-on state to the screen-off state. When proximity sensor 1116 detects that the distance between the user and the front of terminal 1100 is gradually increasing, processor 1101 controls display screen 1105 to switch from the screen-off state to the screen-on state.

[0172] Those skilled in the art will understand that Figure 11 The structure shown in the figure does not constitute a limitation on the terminal 1100, and the terminal 1100 may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0173] Optionally, the computer device is provided as a server. Figure 12 1 is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server 1200 may vary significantly due to different configurations or performance, and may include one or more processors (central processing units, CPUs) 1201 and one or more memories 1202. The memories 1202 store at least one computer program, which is loaded and executed by the processor 1201 to implement the compound property parameter determination methods provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which are not described in detail here.

[0174] An embodiment of the present application further provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed in the compound property parameter determination method of the above embodiment.

[0175] The present application also provides a computer program product or computer program, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the operations performed in the compound property parameter determination method in the various optional implementations described above.

[0176] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0177] The above description is merely an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for determining compound property parameters, characterized in that: The method comprises: Training a prediction model based on at least one first sample data, wherein the first sample data includes structural data and a sample partition coefficient of a first sample compound, wherein the sample partition coefficient is a partition coefficient of the first sample compound between an aqueous phase and a lipid phase, and the trained prediction model is used to determine the partition coefficient of any compound based on the structural data of the compound; retraining the trained prediction model according to at least one second sample data, wherein the second sample data includes structural data and sample attribute parameters of a second sample compound, wherein the sample attribute parameters are attribute parameters related to the partition coefficient; The retrained prediction model is called to determine the target attribute parameters of any compound according to the structural data of any compound.

2. The method according to claim 1, characterized in that The retraining of the trained prediction model according to at least one second sample data includes: For any second sample data, calling the trained prediction model, and determining the predicted attribute parameter of the second sample compound according to the structural data of the second sample compound in the second sample data; The trained prediction model is retrained according to the prediction attribute parameters and the sample attribute parameters in the second sample data.

3. The method according to claim 2, characterized in that The retraining of the trained prediction model according to the prediction attribute parameter and the sample attribute parameter in the second sample data includes: Determining a first loss value of the trained prediction model according to the prediction attribute parameter and the sample attribute parameter in the second sample data; The trained prediction model is retrained according to the first loss value.

4. The method according to claim 1, wherein The trained prediction model includes multiple network layers, and retraining the trained prediction model according to at least one second sample data includes: While keeping the weights of other network layers in the multiple network layers unchanged, adjust the weight of the Kth network layer in the multiple network layers according to any second sample data, where K is an integer not less than 2.

5. The method according to claim 4, characterized in that The method further comprises: Under the condition that the weights of other network layers in the multiple network layers remain unchanged, the weight of the K-1th network layer in the multiple network layers is adjusted according to the next second sample data of any one second sample data.

6. The method according to any one of claims 1 to 5, characterized in that The step of training the prediction model based on at least one first sample data includes: For any piece of first sample data, calling the prediction model, and determining the predicted partition coefficient of the first sample compound according to the structural data of the first sample compound in the first sample data; The prediction model is trained according to the prediction distribution coefficient and the sample distribution coefficient in the first sample data.

7. The method according to claim 6, characterized in that The step of training the prediction model according to the prediction distribution coefficient and the sample distribution coefficient in the first sample data includes: determining a second loss value of the prediction model according to the prediction allocation coefficient and the sample allocation coefficient in the first sample data; The prediction model is trained according to the second loss value.

8. The method according to any one of claims 1 to 5, characterized in that Before training the prediction model based on at least one first sample data, the method further includes: Simulating a motion trajectory of any first sample compound in a mixed solution, wherein the mixed solution is a solution obtained by mixing the first sample compound with a polar solvent and a non-polar solvent; The partition coefficient of the first sample compound when thermodynamic equilibrium is reached between the lipid phase and the aqueous phase of the mixed solution is determined as the sample partition coefficient of the first sample compound, wherein the lipid phase is formed by the non-polar solvent and the first sample compound dissolved in the non-polar solvent, and the aqueous phase is formed by the polar solvent and the first sample compound dissolved in the polar solvent.

9. The method according to any one of claims 1 to 5, characterized in that Before training the prediction model based on at least one first sample data, the method further includes: Obtaining a structure string of any first sample compound, the structure string conforming to the Simplified Molecular Linear Input Specification (SMILES), and determining the structure string as the structure data of the first sample compound; or, A structural diagram of any first sample compound is obtained, and the structural diagram is determined as structural data of the first sample compound.

10. The method according to any one of claims 1 to 5, characterized in that The target property parameters of the compound include the apparent permeability coefficient, solubility in the target property solvent or plasma protein binding rate of the compound.

11. A device for determining compound property parameters, characterized in that: The device comprises: a first training module, configured to train a prediction model based on at least one first sample data, wherein the first sample data includes structural data and a sample partition coefficient of a first sample compound, wherein the sample partition coefficient is the partition coefficient of the first sample compound between an aqueous phase and a lipid phase, and the trained prediction model is configured to determine the partition coefficient of any compound based on the structural data of the compound; a second training module, configured to retrain the trained prediction model based on at least one second sample data, wherein the second sample data includes structural data and sample attribute parameters of a second sample compound, wherein the sample attribute parameters are attribute parameters related to the partition coefficient; The property parameter determination module is used to call the retrained prediction model and determine the target property parameter of any compound according to the structural data of any compound.

12. The device according to claim 11, characterized in that The second training module includes: an attribute parameter prediction unit, configured to call the trained prediction model for any second sample data and determine the predicted attribute parameter of the second sample compound according to the structural data of the second sample compound in the second sample data; The first training unit is used to retrain the trained prediction model according to the prediction attribute parameters and the sample attribute parameters in the second sample data.

13. The device according to claim 12, characterized in that The first training unit is used to determine a first loss value of the trained prediction model based on the prediction attribute parameters and the sample attribute parameters in the second sample data; and retrain the trained prediction model based on the first loss value.

14. A computer device, characterized in that: The computer device includes a processor and a memory, wherein at least one computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the operations performed by the compound property parameter determination method according to any one of claims 1 to 10.

15. A computer-readable storage medium, characterized in that The storage medium stores at least one computer program, which is loaded and executed by a processor to implement the operations performed by the method for determining compound property parameters according to any one of claims 1 to 10.

Images