Protein interaction hotspot prediction method and device

Abstract

The invention discloses a protein interaction hotspot prediction method and device. The method comprises the following steps: obtaining three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule; obtaining surface features corresponding to each sampling point, wherein the surface features comprise materialization attribute features and / or geometric features of each sampling point; based on the three-dimensional coordinates of all the sampling points, a graph structure with the sampling points as nodes and the proximity relation between the sampling points as edges is constructed, and the surface features of all the sampling points are integrated into the graph structure; and using a pre-trained graph neural network model to perform hot spot prediction on the graph structure integrated with the surface features to obtain a protein interaction hot spot prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioinformatics, in particular to a protein interaction hotspot prediction method. The present application also relates to a protein interaction hotspot prediction device, an electronic device and a computer readable storage medium. BACKGROUND

[0002] The prior art has the following problems in predicting protein interaction hotspots: relying on known complex structures or evolutionary information limits its application to target proteins with unknown structures, lack of homologous sequences or binding partner information; the hotspot prediction results lack specific guidance for downstream binder / nanobody design processes, do not provide hotspot information with structural guidance that can be directly used in structure generation processes, and cannot be directly used to improve design success rate and efficiency; the generalization ability of existing protein interaction hotspot prediction methods is insufficient for target points with insufficient structural information in non-model organisms (such as plants, fungi, etc.). SUMMARY

[0003] The present application provides a protein interaction hotspot prediction method, device, electronic device and computer readable storage medium to solve the problems of application limitation, lack of specific guidance for downstream binder / nanobody design processes in hotspot prediction results, and insufficient generalization ability of the prior art in predicting protein interaction hotspots.

[0004] In order to solve or improve the above technical problems to some extent, according to one aspect of the present application, a protein interaction hotspot prediction method is provided, which comprises: obtaining three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule; obtaining surface features corresponding to each sampling point, the surface features including intrinsic features of each sampling point; based on the three-dimensional coordinates of each sampling point, constructing a graph structure with the sampling points as nodes and the adjacent relationship between the sampling points as edges, and integrating the surface features of each sampling point into the graph structure; using a pre-trained graph neural network model to predict hotspots of the graph structure integrated with the surface features, and obtaining a protein interaction hotspot prediction result.

[0005] In one embodiment, the surface features further include associated sequence features of each sampling point, which represent sequence features of amino acids matching the sampling points in distance; The obtaining of the surface features corresponding to each sampling point comprises: obtaining sequence features corresponding to each amino acid in the amino acid sequence of the protein molecule; determine an amino acid matching the sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determine the sequence feature of the amino acid as the associated sequence feature of the corresponding sampling point.

[0006] In an implementation, obtaining the sequence feature corresponding to each amino acid in the amino acid sequence of the protein molecule comprises: inputting the amino acid sequence into a pre-trained protein language model to extract sequence features, and obtaining the sequence context embedding vector of each amino acid residue in the amino acid sequence.

[0007] In an implementation, the surface feature further comprises an associated structure feature of each sampling point, the associated structure feature representing the structure feature of the amino acid matching the sampling point in distance; The obtaining of the surface feature corresponding to each sampling point comprises: obtaining the structure feature corresponding to each amino acid in the amino acid sequence of the protein molecule; determining an amino acid matching the sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determining the structure feature of the amino acid as the associated structure feature of the corresponding sampling point.

[0008] In an implementation, the obtaining of the structure feature corresponding to each amino acid in the amino acid sequence of the protein molecule comprises: inputting the amino acid sequence into a protein structure prediction model to obtain predicted protein three-dimensional structure data; obtaining the structure feature vector corresponding to each amino acid residue based on the protein three-dimensional structure data.

[0009] In an implementation, the surface feature further comprises an associated type feature of each sampling point, the associated structure feature representing the type feature of the amino acid matching the sampling point in distance; The obtaining of the surface feature corresponding to each sampling point comprises: obtaining the type feature of each amino acid in the amino acid sequence of the protein molecule; determining an amino acid matching the sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determining the type feature of the amino acid as the type feature corresponding to the sampling point.

[0010] In an implementation, the inherent feature of the sampling point comprises the physicochemical property feature and / or geometric feature of the surface region of the protein molecule where the sampling point is located.

[0011] In an implementation, the physicochemical property feature comprises at least one of the following: Electrostatic potential; Hydrogen bonding potential; Hydrophobicity.

[0012] In an embodiment, the framework of the graph neural network model comprises a dynamic graph convolutional neural network and a graph attention network.

[0013] According to another aspect of the present application, there is provided a protein interaction hotspot prediction device, comprising: a coordinate obtaining unit configured to obtain three-dimensional coordinates of a plurality of sampling points on a protein molecule surface; a surface feature obtaining unit configured to obtain surface features corresponding to each sampling point, the surface features comprising intrinsic features of each sampling point; a graph structure constructing unit configured to construct a graph structure with the sampling points as nodes and adjacent relationships between the sampling points as edges based on the three-dimensional coordinates of each sampling point, and integrate the surface features of each sampling point into the graph structure; a hotspot predicting unit configured to use a pre-trained graph neural network model to predict hotspots of the graph structure integrated with the surface features, and obtain a protein interaction hotspot prediction result.

[0014] According to another aspect of the present application, there is provided an electronic device comprising a processor and a memory; wherein the memory is configured to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the above method.

[0015] According to another aspect of the present application, there is provided a computer readable storage medium having stored thereon one or more computer instructions, which are executed by a processor to implement the above method.

[0016] Compared with the prior art, the present application has the following advantages: The protein interaction hotspot prediction method provided in this application includes: obtaining the three-dimensional coordinates of multiple sampling points on the surface of a protein molecule; obtaining the surface features corresponding to each sampling point, including the inherent features of each sampling point; constructing a graph structure with sampling points as nodes and proximity relationships between sampling points as edges based on the three-dimensional coordinates of each sampling point, and integrating the surface features of each sampling point into the graph structure; and using a pre-trained graph neural network model to predict hotspots in the graph structure integrating surface features to obtain the protein interaction hotspot prediction results. This method collects the surface features of each sampling point on the protein surface and performs modeling and hotspot prediction based on these features. Since the protein interaction process essentially depends on the surface properties of the protein molecule, the surface features of the sampling points can eliminate the prediction of internal hidden residues, reduce the amount of data and the complexity of feature extraction, thereby improving the accuracy and efficiency of hotspot prediction. Furthermore, since the hotspot prediction process depends on the surface features of the sampling points, this method is also applicable to unresolved protein sequences, expanding the application scope of hotspot prediction. Moreover, because the surface features include the inherent features of each sampling point, the predicted residue hotspots are more in line with downstream design requirements, possessing structural rationality and physical and chemical interpretability. Attached Figure Description

[0017] Figure 1 This is a flowchart of the protein interaction hotspot prediction method provided in the embodiments of this application; Figure 2 This is a unit block diagram of the protein interaction hotspot prediction device provided in the embodiments of this application; Figure 3 This is a schematic diagram of the logical structure of the electronic device provided in the embodiments of this application; Detailed Implementation

[0018] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0019] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0020] It should be understood that, although the terms first, second, etc. can be employed in describing various information in one or more embodiments, the information should not be limited to such terms. These terms are only used to distinguish one category of information from another. For example, without departing from the scope of one or more embodiments, first can be termed second, and, similarly, second can be termed first. Depending on the context, the word "if" as used herein can be interpreted to mean "when" or "in response to determining".

[0021] The term "a plurality of" in the present application can refer to two or more, and "at least one" can refer to one, two or more.

[0022] The term "and / or" in the present application is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, the character " / " in the present application generally represents that the front and rear associated objects have an "or" relationship.

[0023] For the hotspot prediction scenario of protein interaction, in order to more accurately perform hotspot prediction, the present application provides a protein interaction hotspot prediction method, a protein interaction hotspot prediction device corresponding to the method, an electronic device and a computer readable storage medium. The following embodiments are provided to describe the above method, device, electronic device and computer readable storage medium in detail.

[0024] The first embodiment of the present application provides a protein interaction hotspot prediction method, which can be applied to the scene of accurately positioning the core target point, such as the scene of drug design optimization (for example, based on the hotspot prediction result, a high selectivity drug is designed to combine only the hotspot area, and the off-target effect caused by combining the non-hotspot area is avoided), the scene of disease mechanism analysis (for example, positioning pathogenic mutations), the scene of protein interaction regulation (for example, designing a peptide inhibitor targeting PPI hotspot to efficiently block abnormal protein interaction). Figure 1 The flowchart of the method provided by the first embodiment of the present application is as follows: Figure 1 The method provided by the present embodiment is described in detail. The embodiments described below are used to explain the principles of the method, and are not limited to actual use.

[0025] Term explanation: Hotspot refers to a small number of key residues that contribute to the binding free energy in protein interaction, and the mutation (such as alanine mutation) of which will significantly weaken or destroy the interaction; Protein hot spot prediction refers to predicting key amino acid residues in a protein-protein interaction interface that contribute most to binding affinity and stability.

[0026] Protein-protein interaction (PPI) refers to the process of two or more protein molecules combining through non-covalent bonds to perform a specific biological function. Nanobody-antigen interaction (NAI) refers to the specific binding between a nanobody and its target molecule (antigen). Dynamic graph convolutional neural network (DGCNN) is a special graph neural network whose key feature is to dynamically recalculate the graph structure (such as k-nearest neighbor relationship) according to the current node features at each layer, which is suitable for processing irregular data such as point clouds.

[0027] Graph attention network (GAT) is a graph neural network that introduces an attention mechanism, allowing nodes to assign different weights to different neighbors when aggregating neighbor information, thus better capturing important local structures in the graph.

[0028] As shown in Figure 1 The protein interaction hot spot prediction method provided by the embodiment includes the following steps: S101, obtaining three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule.

[0029] This step is used to collect the three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule. In this embodiment, for a single input protein amino acid sequence, the amino acid sequence is input into a protein structure prediction model to obtain predicted protein three-dimensional structure data, and the protein three-dimensional structure data is used as a basis to uniformly sample a set of points (point cloud) on the surface of the protein molecule using a method based on MaSIF (Molecular Surface Interaction Fingerprinting), and the three-dimensional coordinates of each sampling point are obtained accordingly.

[0030] S102, obtaining surface features corresponding to each sampling point.

[0031] Before or after obtaining the three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule in the above step, this step is used to obtain surface features corresponding to each sampling point, which include inherent features of each sampling point. The inherent features represent the characteristics of the protein molecule surface region where the sampling point is located, which are independent of external conditions or have no association with the external environment. In this embodiment, the inherent features of each sampling point include the physicochemical property features and / or geometric features of the protein molecule surface region where the sampling point is located.

[0032] The physicochemical property feature refers to a physical / chemical related feature of the surface of the protein molecule where the sampling point is located. In this embodiment, the physicochemical property feature of the sampling point includes at least one of the following: an electrostatic potential for describing the electrostatic distribution of the sampling point; a hydrogen bond potential for representing the ability of the sampling point to form a hydrogen bond; and a hydrophobicity of the sampling point (which can be assigned according to the hydrophobicity index of the residue corresponding to the sampling point). The geometric feature of the sampling point can be a shape index for describing the geometric shape (concave, convex, flat, etc.) of the local surface where the sampling point is located.

[0033] In another embodiment, the surface feature of each sampling point includes not only the inherent features of the sampling point such as the physicochemical property feature and / or the geometric feature, but also the associated sequence feature of each sampling point, i.e., the surface feature of each sampling point is a fusion feature of the inherent feature and the associated sequence feature, and the associated sequence feature represents the sequence feature of the amino acid matching the sampling point in distance. Correspondingly, the above-mentioned obtaining the surface feature corresponding to each sampling point can refer to: obtaining the sequence feature corresponding to each amino acid in the amino acid sequence of the protein molecule, for example, inputting the amino acid sequence into a pre-trained protein language model to extract the sequence feature, and obtaining the sequence context embedding vector of each amino acid residue in the amino acid sequence; determining the amino acid matching the sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determining the sequence feature of the amino acid as the associated sequence feature of the corresponding sampling point (for example, using the Euclidean distance mapping method, the sequence context embedding vector corresponding to the amino acid closest to the sampling point is taken as the associated sequence feature of the sampling point).

[0034] In another implementation, the surface feature of each sampling point comprises a correlation structure feature of the sampling point in addition to the intrinsic feature of the sampling point such as the physicochemical property feature and / or the geometric feature, i.e., the surface feature of each sampling point is a fusion feature of the intrinsic feature and the correlation structure feature, and the correlation structure feature represents a structure feature of an amino acid that matches the sampling point in distance; correspondingly, the obtaining of the surface feature corresponding to each sampling point can refer to: obtaining a structure feature corresponding to each amino acid in the amino acid sequence of the protein molecule; determining an amino acid that matches each sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determining a structure feature of the amino acid as a correlation structure feature of the corresponding sampling point (for example, using the Euclidean distance mapping method, a structure feature corresponding to an amino acid closest to the sampling point is taken as the correlation structure feature of the sampling point). The obtaining of the structure feature corresponding to each amino acid in the amino acid sequence of the protein molecule can specifically refer to: inputting the amino acid sequence into a protein structure prediction model (for example, a single sequence structure prediction model ESMFold) to obtain predicted protein three-dimensional structure data (for example, atomic coordinates predicted by ESMFold); and obtaining a structure feature vector corresponding to each amino acid residue based on the protein three-dimensional structure data (for example, using the protein secondary structure assignment system (DSSP) to analyze the atomic coordinates predicted by ESMFold to calculate the following structure features of each residue: secondary structure, main chain dihedral angle, solvent accessible surface area).

[0035] In another implementation, the surface feature of each sampling point comprises a correlation type feature of the sampling point in addition to the intrinsic feature of the sampling point such as the physicochemical property feature and / or the geometric feature, i.e., the surface feature of each sampling point is a fusion feature of the intrinsic feature and the correlation type feature, and the correlation type feature represents a type feature of an amino acid that matches the sampling point in distance; correspondingly, the obtaining of the surface feature corresponding to each sampling point can refer to: obtaining a type feature of each amino acid in the amino acid sequence of the protein molecule (for example, performing One-Hot Encoding on the type of each amino acid in the amino acid sequence of the protein molecule to generate a type feature vector); determining an amino acid that matches each sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determining a type feature of the amino acid as a type feature corresponding to the sampling point (for example, using the Euclidean distance mapping method, a type feature of an amino acid closest to the sampling point is taken as the correlation type feature of the sampling point).

[0036] In another implementation, the surface feature can also be a fusion feature of at least two of the intrinsic feature, the correlation structure feature, the correlation sequence feature, and the correlation category feature of each sampling point, which will not be described herein again.

[0037] It should be noted that the above intrinsic features, associated structure features, associated sequence features, and associated category features are fused by feature splicing, and before feature splicing, the high-dimensional features (such as the associated sequence features) are processed by dimension reduction and feature transformation through a multi-layer perception.

[0038] S103, based on the three-dimensional coordinates of each sampling point, a graph structure is constructed with the sampling points as nodes and the adjacent relationship between the sampling points as edges, and the surface features of each sampling point are integrated into the graph structure.

[0039] After obtaining the surface features corresponding to each sampling point in the above step, this step is used to construct a graph structure with the sampling points as nodes and the adjacent relationship between the sampling points as edges based on the three-dimensional coordinates of each sampling point, and integrate the surface features of each sampling point into the graph structure, so as to represent the surface of the protein molecule as a graph structure fused with the surface features of each sampling point.

[0040] In this embodiment, a k-nearest neighbor graph can be constructed based on the three-dimensional coordinates of each sampling point, for example, a (k-NN graph, k=20), that is, each node is connected to its nearest 20 nodes in the Euclidean space to form a set of edges. In another implementation, the graph structure can also be constructed based on a fixed radius neighborhood (ε-ball) or by mixing the use of k-NN and radial basis functions, which will not be described here.

[0041] S104, using a pre-trained graph neural network model to predict hotspots of the graph structure integrated with the surface features, to obtain a protein interaction hotspot prediction result.

[0042] This step is used to input the above graph structure integrated with the surface features of each sampling point into a pre-trained graph neural network model for hotspot prediction, to obtain a protein interaction hotspot prediction result output by the graph neural network model. The protein interaction hotspot prediction result is the probability of each residue becoming a hotspot, but the predicted hotspot is not only a probability value, and the corresponding surface features and physicochemical property features (such as the region being concave, having positive charge) can provide a basis for researchers to make rational judgments and optimize design.

[0043] In the present embodiment, the architecture of the graph neural network model can include a dynamic graph convolutional neural network and a graph attention network. For example, for long-range epitopes commonly seen in nanobody-antigen interaction (NAI) hotspot prediction, the present embodiment additionally introduces a graph attention network (GAT) module based on a dynamic graph convolutional neural network (DGCNN), captures long-range spatial dependencies between surface points through an attention mechanism, can be used to process dynamic and irregular protein surfaces, and can more deeply understand the local geometry and global context of the surface. It should be noted that for protein-protein interaction (PPI) hotspot prediction, the GAT module can be omitted. In the prediction layer, the feature vector of each node of the graph neural network is mapped to a scalar through a multi-layer perceptron (MLP), and an amplification factor (e.g., 5 times) is introduced to enhance the discriminability of the output signal. Finally, the scalar is converted to a probability value between 0 and 1 through a Sigmoid function, representing the probability that the surface point corresponds to a hotspot residue.

[0044] Moreover, in the training phase of the graph neural network model, Focal Loss is used as the loss function to solve the class imbalance problem of hotspot residues (positive samples) being extremely sparse in the data. This function adjusts the weights so that the model pays more attention to difficult-to-classify samples (usually rare positive samples hotspots) during training. At the data level, the graph construction method of the surface center itself filters out a large number of negative samples (buried residues), alleviating the imbalance problem from the data source. That is, the present embodiment naturally excludes buried residues through surface features and combines the Focal Loss loss function for model training, to jointly solve the problem of extremely sparse interaction hotspots in the data.

[0045] The protein interaction hotspot prediction method provided by the embodiment of the present application comprises: obtaining three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule; obtaining surface features corresponding to each sampling point, the surface features comprising inherent features of each sampling point; constructing a graph structure with the sampling points as nodes and the adjacent relationship between the sampling points as edges based on the three-dimensional coordinates of each sampling point, and integrating the surface features of each sampling point into the graph structure; using a pre-trained graph neural network model to perform hotspot prediction on the graph structure integrated with the surface features, and obtaining a protein interaction hotspot prediction result. The method collects the surface features of each sampling point on the surface of the protein, and performs modeling and hotspot prediction based on the surface features. Since the process of protein interaction essentially depends on the surface properties of the protein molecule, the prediction of hidden internal residues can be excluded based on the surface features of the sampling points, the data volume and the complexity of feature extraction are reduced, and the accuracy and efficiency of hotspot prediction are improved. Moreover, since the hotspot prediction process depends on the surface features of the sampling points, the method is also applicable to protein sequences that have not been analyzed, expanding the application range of hotspot prediction. Moreover, since the surface features include inherent features of each sampling point, the predicted residue hotspots are more consistent with the downstream design requirements, and have structural rationality and materialization interpretability.

[0046] For example, the constructed feature naturally excludes internal buried residues, concentrates computing resources on surface areas that are more likely to interact, and the predicted hotspot residues are more likely to be spatially aggregated on the real interaction interface, and their physicochemical properties are more consistent with known PPI or NAI hotspot characteristics, effectively reducing false positives on non-interface regions or buried residues. The output of the present application is the hotspot probability with three-dimensional coordinates, which can be directly input as a spatial constraint condition into a protein design tool based on a diffusion model or an energy function. This kind of "prediction-design" integrated scheme provides clear structural guidance for the generation process, greatly reducing the conformational sampling space. The input features of the model itself contain surface physicochemical properties (such as electrostatic potential, hydrophobicity, hydrogen bond potential) and geometric properties (shape index) with clear physical meaning, therefore, the prediction result of the model is obtained based on such interpretable features, which can improve the structural rationality and materialization interpretability of the hotspot prediction result.

[0047] Moreover, the prior art relies heavily on known protein complex crystal structures (such as HotPoint, KFC2, PredHS2) or evolutionary information provided by multiple sequence alignment (MSA) (such as SPOTONE) for input data requirements, which greatly limits its application on "orphan proteins" or non-model organism proteins with unknown structures or low homology. The present embodiment only requires a single protein amino acid sequence as input, integrates a pre-trained protein language model (e.g., ProtTrans) and a single sequence structure prediction tool (ESMFold) to directly generate the required sequence embedding and three-dimensional structure from the sequence, completely eliminating the dependence on known complex structures or multiple sequence alignment, realizing end-to-end prediction from "sequence -> surface structure -> function (hotspot)", and greatly expanding the applicable scenarios. For example, the method provided by the present embodiment can be successfully applied to plant proteins (NbPDS1) and fungal effector proteins (VdPDA1) and other target points that are traditionally lacking in structural data and interaction information, and effectively guide the design of binding agents and nanobodies for these target points.

[0048] The above embodiment provides a protein interaction hotspot prediction method. Correspondingly, another embodiment of the present application also provides a protein interaction hotspot prediction device. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple. For the details of the related technical features, please refer to the corresponding description of the above-provided method embodiment. The following description of the device embodiment is only illustrative.

[0049] Please refer to Figure 2 To understand this embodiment, Figure 2 The unit block diagram of the protein interaction hotspot prediction device provided by the present embodiment is shown in Figure 2 The protein interaction hotspot prediction device provided by the present embodiment comprises: A coordinate obtaining unit 201 is configured to obtain three-dimensional coordinates of a plurality of sampling points on the surface of a protein molecule. A surface feature obtaining unit 202 is configured to obtain surface features corresponding to each sampling point, wherein the surface features include physicochemical property features and / or geometric features of each sampling point. A graph structure constructing unit 203 is configured to construct a graph structure taking the sampling points as nodes and the adjacent relationship between the sampling points as edges based on the three-dimensional coordinates of the sampling points, and integrate the surface features of the sampling points into the graph structure. A hotspot prediction unit 204 is configured to use a pre-trained graph neural network model to perform hotspot prediction on the graph structure integrated with the surface features, and obtain a protein interaction hotspot prediction result.

[0050] In an embodiment, the surface feature further comprises an associated sequence feature of each sampling point, the associated sequence feature representing a sequence feature of an amino acid matching the sampling point in distance; The obtaining of the surface feature corresponding to each sampling point comprises: The obtaining of the sequence feature corresponding to each amino acid in the amino acid sequence of the protein molecule comprises: The sequence feature of the amino acid matching each sampling point in distance is determined based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and the sequence feature of the amino acid is determined as the associated sequence feature of the corresponding sampling point.

[0051] In an embodiment, the obtaining of the sequence feature corresponding to each amino acid in the amino acid sequence of the protein molecule comprises: The amino acid sequence is input into a pre-trained protein language model for sequence feature extraction to obtain a sequence context embedding vector of each amino acid residue in the amino acid sequence.

[0052] In an embodiment, the surface feature further comprises an associated structure feature of each sampling point, the associated structure feature representing a structure feature of an amino acid matching the sampling point in distance; The obtaining of the surface feature corresponding to each sampling point comprises: The obtaining of the structure feature corresponding to each amino acid in the amino acid sequence of the protein molecule comprises: The structure feature of the amino acid matching each sampling point in distance is determined based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and the structure feature of the amino acid is determined as the associated structure feature of the corresponding sampling point.

[0053] In an embodiment, the obtaining of the structure feature corresponding to each amino acid in the amino acid sequence of the protein molecule comprises: The amino acid sequence is input into a protein structure prediction model to obtain predicted protein three-dimensional structure data; The structure feature vector corresponding to each amino acid residue is calculated based on the protein three-dimensional structure data.

[0054] In an embodiment, the surface feature further comprises an associated type feature of each sampling point, the associated structure feature representing a type feature of an amino acid matching the sampling point in distance; The obtaining of the surface feature corresponding to each sampling point comprises: The obtaining of the type feature of each amino acid in the amino acid sequence of the protein molecule comprises: Determine an amino acid matched with each sampling point in distance based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, and determine the type feature of the amino acid as the type feature corresponding to the sampling point.

[0055] In an implementation, the intrinsic feature of the sampling point includes a physicochemical property feature and / or a geometric feature of a surface region of the protein molecule where the sampling point is located.

[0056] In an implementation, the physicochemical property feature includes at least one of: electrostatic potential; hydrogen bond potential; hydrophobicity.

[0057] In an implementation, the framework of the graph neural network model includes a dynamic graph convolutional neural network and a graph attention network.

[0058] The embodiments of the present application also provide an electronic device which can be provided with the above-mentioned protein interaction hotspot prediction device in the form of a program to execute the protein interaction hotspot prediction method provided by the embodiments of the present application. Optionally, an optional hardware structure of the electronic device can be as shown in the figure, which includes at least one processor 301, at least one communication interface 302, at least one memory 303 and at least one communication bus 304. Figure 3 Optionally, the communication interface 302 can be an interface of a communication module. The processor 301 can be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0059] The memory 303 can include a high-speed RAM memory, and can also include a non-volatile memory such as at least one disk memory.

[0060] The memory 303 stores a program, and the processor 301 invokes the program stored in the memory 303 to execute the protein interaction hotspot prediction method provided by the above-mentioned embodiments of the present application.

[0061] ​In the above embodiments, a method, a device and an electronic device are provided. In addition, another embodiment of the present application also provides a computer readable storage medium for implementing the above method. The computer readable storage medium provided by the present application is described relatively simply, and the related parts can be referred to the corresponding description of the above method embodiments. The embodiments described below are only illustrative. The computer readable storage medium provided by the present application stores computer instructions, which are executed by a processor to implement the following steps: obtain three-dimensional coordinates of a plurality of sampling points on the surface of the protein molecule; obtain surface features corresponding to each sampling point, the surface features including intrinsic features of each sampling point; based on the three-dimensional coordinates of each sampling point, construct a graph structure taking the sampling points as nodes and the adjacent relationship between the sampling points as edges, and integrate the surface features of each sampling point into the graph structure; use a pre-trained graph neural network model to perform hotspot prediction on the graph structure integrated with the surface features, and obtain a protein-protein interaction hotspot prediction result.

[0062] In an embodiment, the surface features further include associated sequence features of each sampling point, the associated sequence features representing sequence features of amino acids matching the sampling points in distance; The obtaining of the surface features corresponding to each sampling point includes: obtaining sequence features corresponding to each amino acid in the amino acid sequence of the protein molecule; based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, determining amino acids matching the sampling points in distance, and determining the sequence features of the amino acids as the associated sequence features of the corresponding sampling points.

[0063] In an embodiment, the obtaining of the sequence features corresponding to each amino acid in the amino acid sequence of the protein molecule includes: inputting the amino acid sequence into a pre-trained protein language model to extract sequence features, and obtaining sequence context embedding vectors of each amino acid residue in the amino acid sequence.

[0064] In an embodiment, the surface features further include associated structure features of each sampling point, the associated structure features representing structure features of amino acids matching the sampling points in distance; The obtaining of the surface features corresponding to each sampling point includes: obtaining structure features corresponding to each amino acid in the amino acid sequence of the protein molecule; determine, based on a positional relationship between each amino acid in the amino acid sequence and each sampling point, an amino acid that matches the sampling point in distance, and determine a structural feature of the amino acid as an associated structural feature of the corresponding sampling point.

[0065] In an implementation, the obtaining of the structural feature corresponding to each amino acid in the amino acid sequence of the protein molecule includes: inputting the amino acid sequence into a protein structure prediction model to obtain predicted protein three-dimensional structure data; calculating a structural feature vector corresponding to each amino acid residue based on the protein three-dimensional structure data.

[0066] In an implementation, the surface feature further includes an associated type feature of each sampling point, and the associated structural feature represents a type feature of an amino acid that matches the sampling point in distance; the obtaining of the surface feature corresponding to each sampling point includes: obtaining a type feature of each amino acid in the amino acid sequence of the protein molecule; determining, based on a positional relationship between each amino acid in the amino acid sequence and each sampling point, an amino acid that matches the sampling point in distance, and determining a type feature of the amino acid as a type feature corresponding to the sampling point.

[0067] In an implementation, the inherent feature of the sampling point includes a physicochemical property feature and / or a geometric feature of a surface region of the protein molecule where the sampling point is located.

[0068] In an implementation, the physicochemical property feature includes at least one of: electrostatic potential; hydrogen bond potential; hydrophobicity.

[0069] In an implementation, the framework of the graph neural network model includes a dynamic graph convolutional neural network and a graph attention network.

[0070] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0071] The memory can include non-persistent memory in computer readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer readable media.

[0072] 1. Computer-readable media includes permanent and non-permanent, removable and non-removable media can be implemented by any method or technology to store information. 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 technologies, read-only compact discs read-only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device. According to the definition in this paper, computer-readable media does not include non-transitory computer-readable media (transitory media), such as modulated data signals and carriers.

[0073] 2. Those skilled in the art should understand that the embodiments of the present application can be provided as a method, system or computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] Although the present application is disclosed with reference to the preferred embodiments above, it is not intended to limit the present application, and any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application should be defined by the scope defined by the claims of the present application.

Claims

1. A method for predicting protein-protein interaction hotspots, characterized in that, include: Obtain the three-dimensional coordinates of multiple sampling points on the surface of a protein molecule; Obtain the surface features corresponding to each sampling point, wherein the surface features include the inherent features of each sampling point; Based on the three-dimensional coordinates of each sampling point, a graph structure is constructed with the sampling points as nodes and the proximity relationships between sampling points as edges, and the surface features of each sampling point are integrated into the graph structure. Hotspot prediction is performed on the graph structure integrating the surface features using a pre-trained graph neural network model to obtain protein interaction hotspot prediction results.

2. The method according to claim 1, characterized in that, The surface features also include the associated sequence features of each sampling point, which characterize the sequence features of amino acids that match the sampling point in distance; The process of obtaining the surface features corresponding to each sampling point includes: Obtain the sequence characteristics corresponding to each amino acid in the amino acid sequence of the protein molecule; Based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, the amino acid that matches each sampling point in distance is determined, and the sequence characteristics of the amino acid are determined as the associated sequence characteristics of the corresponding sampling point.

3. The method according to claim 2, characterized in that, Obtaining the sequence characteristics corresponding to each amino acid in the amino acid sequence of the protein molecule includes: The amino acid sequence is input into a pre-trained protein language model for sequence feature extraction to obtain the sequence context embedding vector of each amino acid residue in the amino acid sequence.

4. The method according to claim 1, characterized in that, The surface features also include the associated structural features of each sampling point, which characterize the structural features of amino acids that match the sampling points in terms of distance; The process of obtaining the surface features corresponding to each sampling point includes: Obtain the structural features corresponding to each amino acid in the amino acid sequence of the protein molecule; Based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, the amino acid that matches each sampling point in distance is determined, and the structural features of the amino acid are determined as the associated structural features of the corresponding sampling point.

5. The method according to claim 4, characterized in that, The process of obtaining the structural features corresponding to each amino acid in the amino acid sequence of the protein molecule includes: The amino acid sequence is input into the protein structure prediction model to obtain the predicted three-dimensional structure data of the protein. Based on the three-dimensional structure data of the protein, the structural feature vectors corresponding to each amino acid residue are calculated.

6. The method according to claim 1, characterized in that, The surface features also include the association type features of each sampling point, and the association structure features characterize the type features of amino acids that match the sampling points in terms of distance; The process of obtaining the surface features corresponding to each sampling point includes: Obtain the type characteristics of each amino acid in the amino acid sequence of the protein molecule; Based on the positional relationship between each amino acid in the amino acid sequence and each sampling point, the amino acid that matches each sampling point in distance is determined, and the type feature of the amino acid is determined as the type feature corresponding to the sampling point.

7. The method according to claim 1, characterized in that, The inherent characteristics of the sampling point include the physicochemical properties and / or geometric features of the protein molecule surface region where the sampling point is located.

8. The method according to claim 1, characterized in that, The architecture of the graph neural network model includes dynamic graph convolutional neural networks and graph attention networks.

9. A protein interaction hotspot prediction device, characterized in that, include: The coordinate acquisition unit is used to obtain the three-dimensional coordinates of multiple sampling points on the surface of a protein molecule; A surface feature acquisition unit is used to acquire the surface features corresponding to each sampling point, wherein the surface features include the inherent features of each sampling point; The graph structure construction unit is used to construct a graph structure with the sampling points as nodes and the proximity relationships between sampling points as edges based on the three-dimensional coordinates of each sampling point, and to integrate the surface features of each sampling point into the graph structure. The hotspot prediction unit is used to predict hotspots on the graph structure integrating the surface features using a pre-trained graph neural network model, and to obtain protein interaction hotspot prediction results.

10. An electronic device, characterized in that, Includes processor and memory; among which, The memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method as described in any one of claims 1-8.

11. A computer-readable storage medium storing one or more computer instructions thereon, characterized in that, The instruction is executed by the processor to implement the method as described in any one of claims 1-8.

Images