Model training method and apparatus, and molecular property information prediction method and apparatus

Abstract

A model training method and apparatus, and a molecular property information prediction method and apparatus. The model training method comprises: acquiring data of a specified protein degradation targeting chimera molecule; on the basis of the data, constructing three-dimensional molecular map data of the specified protein degradation targeting chimera molecule; inputting the three-dimensional molecular map data of the specified protein degradation targeting chimera molecule into a prediction model to be trained, so that said prediction model predicts the molecular property information of the specified protein degradation targeting chimera molecule; and training said prediction model on the basis of a difference between the predicted molecular property information and actual molecular property information corresponding to the specified protein degradation targeting chimera molecule.

Description

A model training method and a method and device for predicting molecular property information Technical Field

[0001] The present application relates to the fields of artificial intelligence and bioengineering, and in particular to a method for model training and a method and device for predicting molecular property information. Background Art

[0002] The ubiquitinase system is a classic pathway for protein degradation in the body. Leveraging this classic pathway, protein degradation-targeted chimeras with bifunctional fragments can be designed to reduce or even eliminate pathogenic proteins in the body. Protein degradation-targeted chimeras are a type of targeted protein degradation technology that utilizes small molecule compounds to regulate protein levels. Compared to traditional small molecule drugs, they have a different mode of action, utilizing unique targets to deliver proteins to the proteasome to directly mediate their ubiquitination and degradation.

[0003] Currently, protein degradation targeted chimera technology is a hot topic in related pharmaceutical research. Such molecules have been used to develop drugs to treat diseases including cancer, immune system diseases, and nervous system diseases. Especially in the field of anti-cancer, protein degradation targeted chimeras can target proteins that induce cancer cells to produce, thereby reducing the side effects of chemotherapy drugs.

[0004] Molecular property prediction is crucial for applications such as drug discovery and protein design. By analyzing a molecular structure model to estimate its relevant physical and chemical properties and calculating accurate molecular properties, the efficiency and accuracy of technology design can be greatly improved. However, there is currently no effective way to accurately predict molecular properties.

[0005] Summary of the Invention

[0006] This application provides a model training method and a molecular structure information recommendation method and device to partially solve the above-mentioned problems existing in the prior art. This application adopts the following technical solutions.

[0007] In a first aspect, the present application provides a method for model training, comprising: obtaining data of a designated protein degradation targeting chimera molecule; constructing three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule based on the data; inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into a prediction model to be trained, so that the prediction model predicts molecular property information of the designated protein degradation targeting chimera molecule; and training the prediction model based on the deviation between the predicted molecular property information and the actual molecular property information corresponding to the designated protein degradation targeting chimera molecule.

[0008] Optionally, the molecular property information includes: the molecular property information includes: at least one of the lipophilicity of the designated protein degradation targeting chimera molecule, the pH of the designated protein degradation targeting chimera molecule, the molecular weight of the designated protein degradation targeting chimera molecule, the hydrogen bond donors and hydrogen bond acceptors in the designated protein degradation targeting chimera molecule, the solubility of the designated protein degradation targeting chimera molecule, and the permeability of the designated protein degradation targeting chimera molecule.

[0009] Optionally, constructing the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule based on the data includes: taking each atom contained in the designated protein degradation targeting chimera molecule as an atomic node, and constructing an adjacency matrix corresponding to each atom to represent the edges between each atomic node, so as to obtain the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule, wherein the adjacency matrix is ​​used to represent the bonding information between the connected atoms.

[0010] Optionally, the node information corresponding to the atomic node of each atom contained in the three-dimensional molecular graph data includes at least one of atom type information, atomic three-dimensional coordinate information, and atomic nuclear charge number.

[0011] Optionally, for each atom contained in the three-dimensional molecular graph data, the atom type information is used to indicate which type of atom the atom specifically belongs to, and is encoded using a one-hot encoding method. The atomic three-dimensional coordinate information is the three-dimensional coordinate information of the atom in the three-dimensional coordinate system determined by projecting the atom into a preset three-dimensional coordinate system, wherein the three-dimensional coordinate system includes a three-dimensional Cartesian coordinate system constructed with the molecular center of mass of the specified protein degradation targeting chimera molecule as the coordinate origin.

[0012] Optionally, the step of inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the prediction model to be trained so that the prediction model predicts the molecular property information of the designated protein degradation targeting chimera molecule includes: inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the embedding layer of the prediction model to obtain an embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer; inputting the embedding vector into the information update layer of the prediction model to determine the target invariant features and target equivariant features of the designated protein degradation targeting chimera molecule; and inputting the final invariant features and the final equivariant features into the output layer of the prediction model to output the predicted molecular property information of the designated protein degradation targeting chimera molecule through the output layer.

[0013] Optionally, the step of inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the embedding layer in the prediction model to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer includes: for each atom in the designated protein degradation targeting chimera molecule, determining the atomic feature information corresponding to the atom based on the three-dimensional molecular graph data, the atomic feature information including at least one of the atomic type information of the atom, the spacing information between the atom and other atoms, and the adjacent atom information of the atom; and inputting the atomic feature information corresponding to each atom into the embedding layer in the prediction model to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer.

[0014] Optionally, the embedding vector includes an initial invariant feature, and the inputting of the embedding vector into the information updating layer in the prediction model to determine the target invariant feature and the target equivariant feature of the specified protein degradation targeting chimera molecule includes: determining the attention weight corresponding to the initial invariant feature through the information updating layer, and determining the attention feature corresponding to the specified protein degradation targeting chimera molecule based on the attention weight; and determining the target invariant feature and the target equivariant feature of the specified protein degradation targeting chimera molecule based on the attention feature through the information updating layer.

[0015] Optionally, the target invariant feature and the target equivariant feature of the specified protein degradation targeting chimera molecule are determined through the information update layer according to the attention feature, including: obtaining the attention feature corresponding to the initial equivariant feature and the attention feature corresponding to the initial invariant feature through the information update layer, and the initial equivariant feature is a zero vector; updating the attention feature corresponding to the initial invariant feature according to the attention feature corresponding to the initial equivariant feature to obtain an updated initial invariant feature; updating the attention feature corresponding to the initial equivariant feature according to the attention feature corresponding to the initial invariant feature to obtain an updated initial equivariant feature; determining the target invariant feature through the information update layer according to the updated initial invariant feature and the attention feature corresponding to the initial invariant feature, and determining the target equivariant feature according to the updated initial equivariant feature, the updated initial invariant feature and the attention feature corresponding to the initial invariant feature.

[0016] Optionally, in response to the number of information update layers being equal to 1, the target equivariant feature and the target invariant feature are respectively used as the final equivariant feature and the final invariant feature; in response to the number of information update layers being greater than 1, the target equivariant feature and the target invariant feature determined by the last information update layer of the prediction model are respectively used as the final equivariant feature and the final invariant feature.

[0017] In a second aspect, the present application provides a method for predicting molecular property information, comprising: obtaining data of a target protein degradation-targeted chimera molecule; constructing three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule based on the data; inputting the three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule into a pre-trained prediction model so that the prediction model predicts the molecular property information of the target protein degradation-targeted chimera molecule, wherein the prediction model is trained by the method described in the first aspect.

[0018] In a third aspect, the present application provides a device for model training, comprising: an acquisition module for acquiring data of a specified protein degradation targeting chimera molecule; a construction module for constructing three-dimensional molecular graph data of the specified protein degradation targeting chimera molecule based on the data; a prediction module for inputting the three-dimensional molecular graph data of the specified protein degradation targeting chimera molecule into a prediction model to be trained, so that the prediction model predicts the molecular property information of the specified protein degradation targeting chimera molecule; and a training module for training the prediction model based on the deviation between the predicted molecular property information and the actual molecular property information corresponding to the specified protein degradation targeting chimera molecule.

[0019] In a fourth aspect, the present application provides a device for predicting molecular property information, comprising: an acquisition module for acquiring data of a target protein degradation-targeted chimera molecule; a construction module for constructing three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule based on the data; and a prediction module for inputting the three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule into a pre-trained prediction model so that the prediction model predicts the molecular property information of the target protein degradation-targeted chimera molecule, wherein the prediction model is trained by the model training method of the first aspect mentioned above.

[0020] In a fifth aspect, the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the model training method of the first aspect mentioned above.

[0021] In a sixth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for predicting molecular property information according to the second aspect.

[0022] In the seventh aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the model training method of the first aspect mentioned above when executing the program.

[0023] In an eighth aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting molecular property information according to the second aspect described above is implemented.

[0024] At least one of the above technical solutions adopted in this application can achieve the following beneficial effects:

[0025] It can be seen from the above method that the present application can construct three-dimensional molecular graph data of the specified protein degradation targeting chimera by obtaining data of the specified protein degradation targeting chimera molecule. This three-dimensional molecular graph data can fully characterize the various characteristics of the molecular structure of the specified protein degradation targeting chimera molecule. Then, after the three-dimensional molecular graph data is input into the prediction model, the prediction model will predict the molecular property information of the specified protein degradation targeting chimera molecule based on the three-dimensional molecular graph data, and then train the prediction model based on the deviation between the predicted molecular property information and the actual molecular property information corresponding to the specified protein degradation targeting chimera molecule, so that in the subsequent process of predicting the molecular property information, the prediction model can be used to quickly and accurately obtain the molecular property information, thereby improving the efficiency and accuracy of determining the molecular property information. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0027] FIG1 is a flow chart of a model training method provided in this application;

[0028] FIG2 is a schematic diagram of a process of predicting molecular property information provided by the present application;

[0029] FIG3 is a schematic diagram of the architecture of a molecular property prediction system provided by the present application;

[0030] FIG4 is a schematic diagram of a model training device provided in this application;

[0031] FIG5 is a schematic diagram of a device for predicting molecular property information provided by the present application;

[0032] FIG6 is a schematic structural diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0033] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0034] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0035] FIG1 is a flow chart of a model training method provided in this application, which includes the following steps:

[0036] S101: Obtain data for a designated protein degradation targeting chimeric molecule.

[0037] The execution entity of the model training method provided in this application can be a terminal device such as a desktop computer, a laptop computer, or a server. For the sake of convenience, this application only uses the terminal device as the execution entity to illustrate the provided model training method.

[0038] In the present application, a terminal device can obtain raw structural data of various protein degradation-targeting chimeric molecules to construct a dataset for training the model. This data can be collected from an external network or obtained through file entry. Exemplary raw data can include SMILES (Simplified Molecular-Input Line-Entry System) datasets of publicly available online protein degradation-targeting chimeric molecules and real protein degradation-targeting chimeric molecules. SMILES datasets are one-dimensional structural representations of compounds.

[0039] After obtaining the above dataset, the terminal device needs to search for data suitable for training the prediction model. In other words, not all data on various protein degradation targeting chimera molecules included in the dataset are suitable as samples for training. Some data may not have well-labeled molecular fragments, and some data may be "dirty data."

[0040] Therefore, the terminal device needs to search for protein degradation targeting chimera molecules suitable as training samples from the data set, that is, to determine the designated protein degradation targeting chimera molecules. A specific implementation method can be to create an extensible three-dimensional molecular graph structure data generator, clean, reconstruct and optimize the protein degradation targeting chimera molecule data, and thus search for the data of the designated protein degradation targeting chimera molecules as training samples, so as to determine the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecules in the subsequent process. The designated protein degradation targeting chimera obtained after cleaning, reconstructing and optimizing the protein degradation targeting chimera compound data has a logP value and PK value associated with the following invariant characteristics and isovariant characteristics, wherein logP represents the oil-water partition coefficient of the compound and the PK value represents the pharmacokinetic quality.

[0041] S102: Constructing three-dimensional molecular graph data of the designated protein degradation targeting chimera based on the data.

[0042] In the present application, after determining the data of the designated protein degradation-targeted chimera molecule from the above-mentioned data set, the terminal device can further determine the three-dimensional molecular graph data of the designated protein degradation-targeted chimera. The three-dimensional molecular graph data mentioned here is used to represent structural features such as the type of each atom in the designated protein degradation-targeted chimera molecule, the position of each atom (such as Cartesian coordinates), the nuclear charge number of each atom, and the connection relationship between each atom.

[0043] For example, each atom contained in a designated protein degradation targeting chimera molecule can be used as an atomic node, and an adjacency matrix corresponding to each atom can be constructed to represent the edges between the atomic nodes, thereby forming part of the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule.

[0044] The adjacency matrix is ​​used to represent the bonding information between connected atoms. For example, for two connected atoms, "0" can be used to represent a single bond connection between the two atoms, "1" can be used to represent a double bond connection, "2" can be used to represent a triple bond connection, and "3" can be used to represent a quadruple bond connection.

[0045] Furthermore, for the above-mentioned three-dimensional molecular graph data, each atom's atomic node corresponds to corresponding node information, and for any atom, the node information of the atom's atomic node can be used to represent some inherent characteristics of the atom. For example, the node information of the atom's atomic node can include: at least one of: atom type information, atomic three-dimensional coordinate information, and atomic nuclear charge number.

[0046] The atom type information is used to indicate the specific type of atom, such as carbon (C), oxygen (O), sulfur (S), nitrogen (N), fluorine (F), or chlorine (Cl). There are many ways to determine the atom type information, such as using one-hot encoding or using predetermined eigenvectors corresponding to the symbols of each element.

[0047] For the atomic three-dimensional coordinate information of each atom, each atom can be projected into a preset three-dimensional coordinate system to determine the three-dimensional coordinate information of each atom in the three-dimensional coordinate system. In some examples, the preset three-dimensional coordinate system can be a three-dimensional Cartesian coordinate system constructed with the molecular center of mass of the designated protein degradation targeting chimeric molecule as the coordinate origin. In other examples, the preset three-dimensional coordinate system can be a three-dimensional coordinate system constructed with the first molecule of the designated protein degradation targeting chimeric molecule as the coordinate origin. This application is not limited to this.

[0048] It can be seen from the above content that due to the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule determined in this application, the structural characteristics of the designated protein degradation targeting chimera molecule can be considered from multiple angles, thereby ensuring the accuracy and rationality of the subsequent prediction model in the prediction results.

[0049] S103: Inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the prediction model to be trained, so that the prediction model predicts the molecular property information of the designated protein degradation targeting chimera molecule.

[0050] In the present application, an embedding layer is provided in the above prediction model, and the input three-dimensional molecular graph data is embedded through the embedding layer to obtain an embedding vector, and then in the subsequent process, the molecular property information is predicted based on the embedding vector.

[0051] In addition, the prediction model in the present application can also be provided with an information update layer. The main function of the information update layer is to input the above-mentioned embedding vector into the information update layer, and then obtain the invariant features and isovariant features of the specified protein degradation targeting chimera molecule through multiple data update operations performed by the information update layer, and then input the invariant features and isovariant features into the output layer of the prediction model to output the predicted molecular property information of the specified protein degradation targeting chimera molecule through the output layer.

[0052] Among them, for each atom in the designated protein degradation targeting chimera molecule, the terminal device can first determine the atomic feature information corresponding to the atom based on the three-dimensional molecular graph data. The atomic feature information mentioned here is used to represent the characteristics of the atom, and can specifically include the atomic type information of the atom (that is, the atomic type information is the information mentioned above for indicating which type of atom it belongs to), the spacing information between the atom and other atoms, and the adjacent atom information of the atom, etc.

[0053] In the present application, the spacing information between two atoms can be determined in a variety of ways. For example, after determining the three-dimensional coordinate information of each atom, the distance between the three-dimensional coordinates of the two atoms can be determined, and the determined distance can be directly used as the spacing information between the two atoms. For another example, the spacing information between the two atoms can be determined by a radial basis function (RBF). Specifically, the spacing information can be determined by using the radial basis function method with reference to the following formula (1):

[0054] Among them, d ij Used to represent the distance between atoms i and j, d c Indicates the distance from the cutoff point. The specific value can be determined according to the requirements. When d ij Equal to d c When , the sin function is equal to 0. e RBF (d ij ) is used to represent the distance information between atoms i and j determined by the radial basis function.

[0055] As for the adjacent atom information mentioned above, it can be determined by the following embedding formula (2):

[0056] embed1 is used to represent the Cauchy graph embedding function. Used to represent linear mapping function, A ij It is used to represent the molecular adjacency matrix containing bonding information, L1 is used to represent the linear transformation function, and e ij It is used to represent the adjacent atom information of atom i and atom j.

[0057] After determining the atomic feature information corresponding to each atom, the atomic feature information corresponding to each atom can be input into the embedding layer in the prediction model to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer. Exemplarily, the embedding vector can include the initial invariant feature shown in the subsequent formula (3-1).

[0058] As mentioned above, in this application, the information update layer is primarily used to input the embedding vector into the information update layer, and then perform multiple data update operations to ultimately obtain the invariant features and equivariant features of the protein degradation targeting chimera molecule. In this application, the information update layer primarily uses an attention mechanism and a message passing mechanism to continuously update the invariant features and equivariant features. After completing the feature update, the updated features are used to predict molecular property information.

[0059] Specifically, after the embedding vector is input into the information update layer of the prediction model, the information update layer can determine the attention weight corresponding to the initial invariant feature, and based on the attention weight, the attention feature corresponding to the initial invariant feature of the specified protein degradation targeting chimera molecule can be determined. Similarly, after the embedding vector is input into the information update layer of the prediction model, the information update layer can determine the attention weight corresponding to the initial equivariant feature, and based on the attention weight, the attention feature corresponding to the initial equivariant feature of the specified protein degradation targeting chimera molecule can be determined. This process can be considered to be completed entirely in the information update layer.

[0060] When determining the initial invariant characteristics of a given protein degradation targeting chimeric molecule, the following formula (3-1) can be used to determine:

[0061] Among them, c i Used to represent the nuclear charge number of atom i, x i It is used to represent the atomic type information of atom i, embed2 is used to represent the Laplace characteristic embedding function, Used to represent linear mapping functions, and is used to represent the initial invariant characteristics of atom i.

[0062] As for the initial equivariant characteristics, they can be determined by initializing the values. For details, please refer to the following formula (3-2):

[0063] From the above formula, we can see that the initial equivariant feature can be initialized as That is the zero vector.

[0064] After determining the initial equivariant feature and the initial invariant feature, the two features can be updated respectively to obtain the target equivariant feature and the target invariant feature.

[0065] The prediction model can first obtain query (Q) and key (K) embeddings through the attention mechanism in the information update layer, which can be determined by the following formulas (4-1) and (4-2): Q = W Q S+bQ (4-1) K=W K S+b k (4-2)

[0066] Among them, W Q The weight coefficient used to represent the attention weight Q corresponding to the invariant feature, b Q The bias coefficient used to represent the attention weight Q corresponding to the invariant feature, W K The weight coefficient used to represent the attention weight K corresponding to the invariant feature, b k The bias coefficient used to represent the attention weight K corresponding to the invariant feature, S represents the combination of invariant features. Initially, the invariant feature can be the invariant feature

[0067] Then, through the attention mechanism in the information update layer, the attention features corresponding to the specified protein degradation targeting chimera molecule can be further determined based on the determined attention weights. For details, please refer to the following formulas (5-1) and (5-2):

[0068] Among them, W a is the weight matrix, Q i Used to represent the attention weights Q, K corresponding to the invariant features of the determined atom i i Used to represent the attention weight K corresponding to the invariant features of the determined atom i, e i Used to represent the adjacent atom information of atom i, e i is the e in the above formula (2) ij A vector composed of each element, each element is the j-th atom and the i-th e ij , y1, y2, y3 have no particular practical meaning and can be regarded as A output The output matrix is ​​split into three matrices, y1, y2, and y3. This representation is mainly used to determine the invariant and equivariant features of atoms in the designated protein degradation targeting chimera molecule in the subsequent process. Ω, It is used to represent the linear transformation function, through which y1 is processed, mainly to convert y1 in the obtained attention feature into a single feature vector.

[0069] For the initial equivariant features, this application can use the VN-MLP equivariant multilayer perceptron mechanism to update them. For details, please refer to the following formulas (6-1) and (6-2):

[0070] Among them, W is used to represent the attention weight Q corresponding to the equivariant featureV The weight coefficient, b1 is used to represent the attention weight Q corresponding to the equivariant feature V Bias coefficient, U is used to represent the attention weight K corresponding to the equivariant feature V The weight coefficient, b2 is used to represent the attention weight K corresponding to the equivariant feature V The bias coefficient of . Initially, Can include equivariant features

[0071] Then, the attention feature for the initial equivariant feature can be determined by the attention weight corresponding to the determined equivariant feature. For details, please refer to the following formula (7):

[0072] After obtaining the attention features for the initial invariant features (also referred to as invariant features) and the attention features for the initial equivariant features (also referred to as equivariant features) in the above manner, the invariant features and equivariant features can be further updated through the information update layer.

[0073] Specifically, through the information update layer, the invariant feature can be updated according to the equivariant feature to obtain the updated invariant feature, wherein the attention feature corresponding to the initial equivariant feature can be first determined by the above method. Then, according to the attention features corresponding to the initial equivariant features To update the invariant feature Si, the following formulas (8-1) and (8-2) can be used:

[0074] Among them, S′ i and S" i It can be regarded as using different equivariant functions to deal with the invariant feature S i The invariant features after two different updates, VN-MLP1 and VN-MLP2 represent the equivariant multilayer perceptron functions. and represents two different equivariant functions.

[0075] As for the update of equivariant features, the equivariant features can be updated through the information update layer according to the invariant features to obtain the updated equivariant features. Specifically, this process can be to first determine the attention features corresponding to the initial equivariant features according to the initial equivariant features, and then, according to the invariant features Si, determine the attention features corresponding to the initial equivariant features. For update, please refer to the following formula (8-3):

[0076] It is used to represent an equivariant function, and diag is used to represent a function that converts a matrix into a diagonal matrix form. Used to represent the updated equivariant features, which is the attention feature corresponding to the initial equivariant features Joint invariant feature S i After updating, VN-MLP3 represents the equivariant multilayer perceptron function.

[0077] After obtaining the updated invariant features, the updated invariant features S′ can be obtained through the information update layer. i And the attention features corresponding to the determined initial invariant features To determine the target invariant features, please refer to the following formula (9-1):

[0078] Among them, ker1 is used to represent the Kelvin function, is used to represent the distance between atoms i and j, and It is used to represent the determined target invariant features.

[0079] For equivariant features, the updated equivariant features can be used Updated invariant feature S" i And the determined attention features y2 and y3 are used to determine the target equivariant features. For details, please refer to the following formula (9-2):

[0080] Among them, ker2 and ker3 are used to represent Kelvin functions. is used to represent the distance between atoms i and j, and It is used to represent the determined target equivariant features.

[0081] It should be noted that the target equivariant features and target invariant features obtained here are only the equivariant features and invariant features output by this information update layer. In some examples, in response to the number of information update layers being equal to 1, they can be directly output to the output layer of the model as the final equivariant features and invariant features. In some examples, in response to the number of information update layers being greater than 1, they can also be output to another information update layer for further updating and optimization of these features.

[0082] In this application, the prediction model may also be provided with multiple information update layers, and the multiple information update layers are used to transfer features to achieve continuous updating and optimization of invariant features and equivariant features, so as to finally input the updated and optimized invariant features and equivariant features into the output layer of the prediction model.

[0083] Therefore, after determining the above-mentioned target invariant features and target equivariant features, they can be combined into a feature combination as shown in the following formula (10-1).

[0084] Then, the feature combination and the original input are input to the next information update layer to obtain the output result of the information update layer, and the output result is passed to the next information update layer, and so on. For details, please refer to the following formula (10-2):

[0085] In this formula, m i That is, the feature combination composed of the target invariant feature of atom i and the target equivariant feature of atom i, X i is the atomic characteristic information of atom i, f a Used to indicate the attention mechanism of m i To process, It is used to represent the output result of the information update layer. It should be pointed out that when determining the target invariant feature of atom i, the atomic feature information of atom j that has a connection relationship with atom i is actually considered. Atom i may not be connected to only one atom j. Therefore, when determining the target invariant feature of atom i, the information between atom i and each connected atom j can actually be considered. Summarize and process the target equivariant features of atom i in the same way, thus obtaining the feature combination m of atom i i .

[0086] The final information update layer of the prediction model outputs the target isovariant and invariant features as the final isovariant and invariant features, which are then fed to the model's output layer. The output layer then combines these final invariant and isovariant features to predict the druggability of the molecule, ultimately obtaining the molecular properties of the designated protein degradation-targeting chimera.

[0087] From the above content, it can be seen that the invariant features of the designated protein degradation targeting chimera molecule determined actually characterize the molecular features of the designated protein degradation targeting chimera molecule that will not change with the molecular structure (because the invariant features of each atom are actually determined by information such as the atomic charge number, and given the molecular structure, this information is usually fixed), while the equivariant features of the designated protein degradation targeting chimera can characterize the characteristics exhibited by some more noteworthy atoms in the molecular structure of the designated protein degradation targeting chimera molecule (because the equivariant features of each atom are actually determined by information such as the three-dimensional coordinate information of each atom, in order to simplify the calculation, the initial equivariant features are set to 0, and the equivariant features are determined through the attention mechanism).

[0088] Therefore, it can be understood that the subsequent output layer actually predicts the molecular property information of the specified protein degradation targeting chimera molecule based on the final isovariant characteristics and invariant characteristics of the specified protein degradation targeting chimera molecule, the characteristics of some more noteworthy atoms in the molecular structure of the specified protein degradation targeting chimera molecule, and some fixed characteristics of the molecular structure of the specified protein degradation targeting chimera molecule.

[0089] S104: Training the prediction model according to the deviation between the predicted molecular property information and the actual molecular property information corresponding to the designated protein degradation targeting chimera molecule.

[0090] After predicting the above molecular property information, the deviation between the molecular property information and the actual molecular property information corresponding to the designated protein degradation targeting chimera molecule can be further determined, and then the prediction model is trained with minimizing the deviation as the optimization goal.

[0091] Among them, the molecular property information mentioned in the present application may include: the lipophilicity of the specified protein degradation targeting chimera molecule, the pH of the specified protein degradation targeting chimera molecule, the molecular weight of the specified protein degradation targeting chimera molecule, the hydrogen bond donors and hydrogen bond acceptors in the specified protein degradation targeting chimera molecule, the solubility of the specified protein degradation targeting chimera molecule, and the permeability of the specified protein degradation targeting chimera molecule.

[0092] As can be seen from the above method, the model training method provided in this application references the structural features of the molecular structure of the designated protein degradation-targeted chimera molecule from multiple angles during the process of determining the three-dimensional molecular graph data of the designated protein degradation-targeted chimera molecule. Furthermore, when predicting the molecular property information of the designated protein degradation-targeted chimera molecule, since the prediction is based on the final invariant and isovariant features of the designated protein degradation-targeted chimera molecule, this method can fully characterize the molecular structural characteristics of the designated protein degradation-targeted chimera molecule, thereby ensuring that the prediction model can subsequently perform accurate and reasonable predictions of molecular property information.

[0093] After training the above prediction model, you can use it to predict the molecular properties of a specific molecule in real-world applications. The specific process is shown in the figure below.

[0094] FIG2 is a schematic diagram of the process of a method for predicting molecular property information provided in this application.

[0095] S201: Acquire data of chimeric molecules targeting degradation of target proteins.

[0096] S202: Constructing three-dimensional molecular graph data of the target protein degradation targeting chimera molecule based on the data.

[0097] In the present application, the terminal device can receive the molecular property prediction instruction of the target protein degradation targeted chimera molecule input by the user, so as to obtain the data of the target protein degradation targeted chimera molecule through the molecular property prediction instruction, and construct the three-dimensional molecular graph data of the target protein degradation targeted chimera molecule. Among them, the molecular property prediction instruction of the target protein degradation targeted chimera molecule can include two fragment molecules provided by the user. The determination of the three-dimensional molecular graph data here is basically the same as the determination process of the three-dimensional molecular graph data in the above-mentioned model training, and will not be described in detail here. The terminal device mentioned here can refer to a desktop computer, a laptop computer and other devices.

[0098] S203: Inputting the three-dimensional molecular graph data of the target protein degradation targeting chimera molecule into a pre-trained prediction model so that the prediction model predicts the molecular property information of the target protein degradation targeting chimera molecule, wherein the prediction model is trained by the above-mentioned model training method.

[0099] The terminal device can input the three-dimensional molecular map data of the target protein degradation targeting chimera molecule into the prediction model deployed in the terminal device, and the prediction model will output the predicted molecular property information of the target protein degradation targeting chimera molecule.

[0100] After the molecular property information is obtained through the prediction model, it can be displayed to the user, or information recommendations can be made to the user based on the molecular property information, such as drug compounds that can be well integrated with the target protein degradation targeted chimera molecule. Of course, in actual applications, the three-dimensional molecular graph data of the target protein degradation targeted chimera molecule can also be stored in association with the predicted molecular property information.

[0101] The present application also provides a system for predicting molecular properties, as shown in FIG3 .

[0102] FIG3 is a schematic diagram of the architecture of a molecular property prediction system provided in this application.

[0103] As can be seen in Figure 3, the system primarily consists of the following components: a storage subsystem for storing the aforementioned datasets, as well as information on molecular properties and their pharmaceutical and chemical properties predicted by the prediction model in actual applications; and a control subsystem for predicting the molecular properties of the target protein degradation-targeted chimera molecule based on the three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule input into the subsystem. The control subsystem includes two units: a molecular feature extraction unit and a molecular property prediction unit. These two units are respectively used to determine the atomic feature information of each atom from the three-dimensional molecular graph data and predict the molecular property information of the target protein degradation-targeted chimera molecule.

[0104] The above are one or more implementation methods of this application. Based on the same idea, this application also provides corresponding model training devices and molecular structure information recommendation devices, as shown in Figures 4 and 5.

[0105] Figure 4 is a schematic diagram of a model training device provided in the present application, including: an acquisition module 401, used to acquire data of a specified protein degradation targeting chimera molecule; a construction module 402, used to construct three-dimensional molecular graph data of the specified protein degradation targeting chimera molecule based on the data; a prediction module 403, used to input the three-dimensional molecular graph data of the specified protein degradation targeting chimera molecule into the prediction model to be trained, so that the prediction model predicts the molecular property information of the specified protein degradation targeting chimera molecule; a training module 404, used to train the prediction model based on the deviation between the predicted molecular property information and the actual molecular property information corresponding to the specified protein degradation targeting chimera molecule.

[0106] Optionally, the molecular property information includes: the molecular property information includes: at least one of the lipophilicity of the designated protein degradation targeting chimera molecule, the pH of the designated protein degradation targeting chimera molecule, the molecular weight of the designated protein degradation targeting chimera molecule, the hydrogen bond donors and hydrogen bond acceptors in the designated protein degradation targeting chimera molecule, the solubility of the designated protein degradation targeting chimera molecule, and the permeability of the designated protein degradation targeting chimera molecule.

[0107] Optionally, the construction module 402 is specifically used to take each atom contained in the specified protein degradation targeting chimera molecule as an atomic node, and to construct an adjacency matrix corresponding to each atom to represent the edges between each atomic node, so as to obtain the three-dimensional molecular graph data of the specified protein degradation targeting chimera molecule, wherein the adjacency matrix is ​​used to represent the bonding information between the connected atoms.

[0108] Optionally, the node information corresponding to the atomic node of each atom contained in the three-dimensional molecular graph data includes at least one of atom type information, atomic three-dimensional coordinate information, and atomic nuclear charge number.

[0109] Optionally, for each atom contained in the three-dimensional molecular graph data, the atom type information is used to indicate which type of atom the atom specifically belongs to, and is encoded using a one-hot encoding method. The atomic three-dimensional coordinate information is the three-dimensional coordinate information of the atom in the three-dimensional coordinate system determined by projecting the atom into a preset three-dimensional coordinate system, wherein the three-dimensional coordinate system includes a three-dimensional Cartesian coordinate system constructed with the molecular center of mass of the specified protein degradation targeting chimera molecule as the coordinate origin.

[0110] Optionally, the prediction module 403 is specifically used to input the three-dimensional molecular graph data of the specified protein degradation targeting chimera molecule into the embedding layer in the prediction model to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer; input the embedding vector into the information update layer in the prediction model to determine the target invariant features and target isovariant features of the specified protein degradation targeting chimera molecule; input the final invariant features and final isovariant features into the output layer in the prediction model to output the predicted molecular property information of the specified protein degradation targeting chimera molecule through the output layer.

[0111] Optionally, the prediction module 403 is specifically used to determine, for each atom in the specified protein degradation targeting chimera molecule, the atomic feature information corresponding to the atom based on the three-dimensional molecular graph data, the atomic feature information including: at least one of the atomic type information of the atom, the spacing information between the atom and other atoms, and the adjacent atom information of the atom; and input the atomic feature information corresponding to each atom into the embedding layer in the prediction model to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer.

[0112] Optionally, the embedding vector includes an initial invariant feature, and the prediction module 403 is specifically used to determine the attention weight corresponding to the initial invariant feature through the information update layer, and determine the attention feature corresponding to the specified protein degradation targeting chimera molecule based on the attention weight; through the information update layer, determine the target invariant feature and the target equivariant feature of the specified protein degradation targeting chimera molecule based on the attention feature.

[0113] Optionally, the prediction module 403 is specifically used to obtain, through the information update layer, an attention feature corresponding to the initial equivariant feature and an attention feature corresponding to the initial invariant feature, wherein the initial equivariant feature is a zero vector; according to the attention feature corresponding to the initial equivariant feature, the attention feature corresponding to the initial invariant feature is updated to obtain an updated initial invariant feature; according to the attention feature corresponding to the initial invariant feature, the attention feature corresponding to the initial equivariant feature is updated to obtain an updated initial equivariant feature; through the information update layer, according to the updated initial invariant feature and the attention feature corresponding to the initial invariant feature, the target invariant feature is determined, and according to the updated initial equivariant feature, the updated initial invariant feature and the attention feature corresponding to the initial invariant feature, the target equivariant feature is determined.

[0114] Optionally, the prediction module 403 is specifically used to, in response to the number of information update layers being equal to 1, use the target equivariant feature and the target invariant feature as the final equivariant feature and the final invariant feature, respectively; in response to the number of information update layers being greater than 1, use the target equivariant feature and the target invariant feature determined by the last information update layer of the prediction model as the final equivariant feature and the final invariant feature, respectively.

[0115] Figure 5 is a schematic diagram of a molecular property information prediction device provided in the present application, including: an acquisition module 501, used to obtain data of a target protein degradation-targeted chimera molecule; a construction module 502, used to construct three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule based on the data; a prediction module 503, used to input the three-dimensional molecular graph data of the target protein degradation-targeted chimera molecule into a pre-trained prediction model so that the prediction model predicts the molecular property information of the target protein degradation-targeted chimera molecule, and the prediction model is trained using the above-mentioned model training method.

[0116] The present application also provides a computer-readable storage medium storing a computer program that can be used to execute the model training method provided in FIG1 or the molecular property information prediction method provided in FIG2. The computer-readable storage medium can be a non-volatile storage medium.

[0117] This application also provides a schematic structural diagram of an electronic device corresponding to Figure 1 or Figure 2, as shown in Figure 6. As shown in Figure 6, the electronic device includes a processor, memory, and non-volatile memory, and may also include hardware required for other services, such as a network interface and an internal bus. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the model training method described in Figure 1 or the molecular property information prediction method described in Figure 2.

[0118] Of course, in addition to software implementation, this application does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0119] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0120] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0121] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0122] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0123] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0124] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0125] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0127] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0128] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0129] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0130] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0131] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0133] The various embodiments in this application are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment.

[0134] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A model training method, It is characterized in that include: Obtain data for a designated protein degradation targeting chimeric molecule; Based on the data, construct three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule; Inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the prediction model to be trained, so that the prediction model predicts the molecular property information of the designated protein degradation targeting chimera molecule; The prediction model is trained according to the deviation between the predicted molecular property information and the actual molecular property information corresponding to the designated protein degradation targeting chimera molecule.

2. The method according to claim 1, It is characterized in that The molecular property information includes at least one of the lipophilicity of the designated protein degradation targeting chimera molecule, the pH value of the designated protein degradation targeting chimera molecule, the molecular weight of the designated protein degradation targeting chimera molecule, the hydrogen bond donors and hydrogen bond acceptors in the designated protein degradation targeting chimera molecule, the solubility of the designated protein degradation targeting chimera molecule, and the permeability of the designated protein degradation targeting chimera molecule.

3. The method according to claim 1, It is characterized in that The constructing of the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule according to the data comprises: Each atom contained in the designated protein degradation targeted chimera molecule is used as an atomic node, and an adjacency matrix corresponding to each atom is constructed to represent the edges between the atomic nodes, so as to obtain the three-dimensional molecular graph data of the designated protein degradation targeted chimera molecule, wherein the adjacency matrix is ​​used to represent the bonding information between the connected atoms.

4. The method according to claim 3, It is characterized in that The node information corresponding to the atomic node of each atom contained in the three-dimensional molecular graph data includes at least one of atom type information, atomic three-dimensional coordinate information, and atomic nuclear charge number.

5. The method according to claim 4, It is characterized in that For each atom contained in the three-dimensional molecular graph data, The atom type information is used to indicate which type of atom the atom belongs to, and is encoded using a one-hot encoding method. The atomic three-dimensional coordinate information is the three-dimensional coordinate information of the atom in the three-dimensional coordinate system determined by projecting the atom into a preset three-dimensional coordinate system, wherein the three-dimensional coordinate system includes a three-dimensional Cartesian coordinate system constructed with the molecular center of mass of the specified protein degradation targeting chimera molecule as the coordinate origin.

6. The method according to claim 1, It is characterized in that The step of inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the prediction model to be trained so that the prediction model predicts the molecular property information of the designated protein degradation targeting chimera molecule comprises: The three-dimensional molecular graph data of the specified protein degradation targeting chimeric molecule is input into the prediction model The embedding layer in , so as to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer; Inputting the embedding vector into an information updating layer in the prediction model to determine the target invariant features and target isovariant features of the designated protein degradation targeting chimera molecule; The final invariant features and the final isovariant features are input into the output layer of the prediction model, so as to output the predicted molecular property information of the designated protein degradation targeting chimera molecule through the output layer.

7. The method according to claim 6, It is characterized in that The step of inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the embedding layer in the prediction model to obtain an embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer comprises: For each atom in the designated protein degradation targeting chimera molecule, determining atomic feature information corresponding to the atom according to the three-dimensional molecular graph data, wherein the atomic feature information includes at least one of atomic type information of the atom, spacing information between the atom and other atoms, and adjacent atomic information of the atom; The atomic feature information corresponding to each atom is input into the embedding layer in the prediction model, so as to obtain the embedding vector corresponding to the three-dimensional molecular graph data through the embedding layer.

8. The method according to claim 6, It is characterized in that The embedding vector includes an initial invariant feature, and the embedding vector is input into the information updating layer in the prediction model to determine the target invariant feature and the target isovariant feature of the designated protein degradation targeting chimera molecule, including: Determining the attention weight corresponding to the initial invariant feature through the information updating layer, and determining the attention feature corresponding to the designated protein degradation targeting chimera molecule according to the attention weight; The target invariant features and the target isovariant features of the designated protein degradation targeting chimera molecule are determined through the information updating layer according to the attention features.

9. The method according to claim 8, It is characterized in that The step of determining the target invariant feature and the target isovariant feature of the designated protein degradation targeting chimera molecule according to the attention feature through the information updating layer includes: Obtaining, through the information updating layer, an attention feature corresponding to an initial equivariant feature and an attention feature corresponding to the initial invariant feature, wherein the initial equivariant feature is a zero vector; According to the attention feature corresponding to the initial equivariant feature, the attention feature corresponding to the initial invariant feature is updated to obtain an updated initial invariant feature; According to the attention feature corresponding to the initial invariant feature, the attention feature corresponding to the initial equivariant feature is updated to obtain an updated initial equivariant feature; The information update layer determines the target invariant feature according to the updated initial invariant feature and the attention feature corresponding to the initial invariant feature, and determines the target invariant feature according to the updated initial equivariant feature, ... target invariant feature, the target The initial invariant feature and the attention feature corresponding to the initial invariant feature determine the target equivariant feature.

10. The method according to claim 6, It is characterized in that In response to the number of the information update layers being equal to 1, using the target equivariant feature and the target invariant feature as the final equivariant feature and the final invariant feature, respectively; In response to the number of the information update layers being greater than 1, the target equivariant feature and the target invariant feature determined by the last information update layer of the prediction model are used as the final equivariant feature and the final invariant feature, respectively.

11. A method for predicting molecular property information. It is characterized in that include: Obtain data on chimeric molecules targeting degradation of target proteins; Based on the data, constructing three-dimensional molecular graph data of the target protein degradation targeting chimera molecule; The three-dimensional molecular graph data of the target protein degradation targeted chimera molecule is input into a pre-trained prediction model so that the prediction model predicts the molecular property information of the target protein degradation targeted chimera molecule, and the prediction model is trained by the method described in any one of claims 1-10 above.

12. A device for model training, It is characterized in that include: An acquisition module, used to acquire data of a designated protein degradation targeting chimera molecule; A construction module, for constructing three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule according to the data; A prediction module, used for inputting the three-dimensional molecular graph data of the designated protein degradation targeting chimera molecule into the prediction model to be trained, so that the prediction model predicts the molecular property information of the designated protein degradation targeting chimera molecule; The training module is used to train the prediction model according to the deviation between the predicted molecular property information and the actual molecular property information corresponding to the designated protein degradation targeting chimera molecule.

13. A device for predicting molecular property information, It is characterized in that include: An acquisition module, used to acquire data of a target protein degradation targeting chimera molecule; A construction module, used to construct three-dimensional molecular graph data of the target protein degradation targeting chimera molecule according to the data; A prediction module is used to input the three-dimensional molecular graph data of the target protein degradation targeted chimera molecule into a pre-trained prediction model so that the prediction model predicts the molecular property information of the target protein degradation targeted chimera molecule, and the prediction model is trained by the method described in any one of claims 1 to 10.

14. A computer-readable storage medium, It is characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

15. A computer-readable storage medium, It is characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to claim 11 is implemented.

16. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the method according to any one of claims 1 to 10 is implemented.

17. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the method according to claim 11 is implemented.

Images