Drug-target interaction prediction method based on geometric representation

Abstract

This invention provides a drug-target interaction method based on geometric representation, comprising: obtaining drug structural features from a drug structure map using a geometrically enhanced graph neural network (GeoGNN); obtaining target structural features from a target structure map using a geometric encoder; constructing a heterogeneous network by combining drug comprehensive similarity, target comprehensive similarity, and drug-target correlation matrix; extracting features from the heterogeneous network using a graph neural network (GCN) to obtain drug network features and target network features; fusing drug structural features and drug network features, and target structural features and target network features, respectively, using multimodal feature fusion; extracting interaction information between the drug and target from the fused drug multimodal features and the fused target multimodal features; obtaining better drug and target features based on the interaction information; and predicting the interaction probability between the drug and target using a classifier. This invention improves the accuracy of drug-target interaction prediction.

Description

Technical Field

[0001] This invention relates to the field of drug target interaction prediction technology, and in particular to a method for predicting drug target interactions based on geometric characterization. Background Technology

[0002] Drug-target interaction (DTI) prediction is a technique that uses computational methods and models to predict the interactions between drugs and biological targets. DTI prediction helps researchers gain a deeper understanding of drug mechanisms of action, discover new drug targets, and screen for potential drug candidates.

[0003] In recent years, with the development of deep learning technology, multimodal features based on sequence, structure, and network characteristics have been widely applied to drug detection and treatment (DTI) prediction. Currently, many DTI prediction models have incorporated features from multiple modalities. These features include, but are not limited to, the chemical structure, sequence information, and related biological network characteristics of drugs or targets. Through deep learning models, researchers can extract useful information from complex data, thereby improving the accuracy and reliability of DTI prediction.

[0004] While existing DTI prediction models can extract features using structural information from drugs or targets, these models typically represent the molecular topology only through atomic properties and interatomic connections, failing to adequately capture complex three-dimensional geometry and the spatial arrangement of atoms within the drug. Furthermore, existing multimodal feature fusion methods often simply stitch together features from multiple modalities or fuse them by introducing learnable weight parameters, without considering how to effectively integrate these multimodal features to further improve prediction performance. Therefore, developing a novel DTI prediction method that can effectively fuse multimodal features and fully consider 3D geometric information and atomic spatial arrangement is particularly important. Summary of the Invention

[0005] In view of this, the present invention provides a drug target interaction prediction method based on geometric characterization to improve the accuracy of drug target interaction prediction.

[0006] Therefore, the present invention provides the following technical solution:

[0007] This invention provides a drug target interaction method based on geometric characterization, the method comprising:

[0008] Drug structure maps and drug fingerprints are obtained from the SMILES sequences of the drug; drug structural features are obtained from the drug structure maps using a geometrically enhanced graph neural network (GeoGNN).

[0009] Obtain the target structure map from the Fasta sequence of the target; use a geometric encoder to obtain the target structure features from the target structure map;

[0010] Calculate the drug fingerprint similarity of the drug and the target sequence similarity of the target separately;

[0011] The Gaussian similarity of drugs and targets is obtained through the drug-target correlation matrix;

[0012] The drug fingerprint similarity and the drug Gaussian similarity are fused to obtain the comprehensive drug similarity;

[0013] The target sequence similarity and the target Gaussian similarity are fused to obtain the comprehensive target similarity.

[0014] The comprehensive similarity of the drugs, the comprehensive similarity of the targets, and the drug-target correlation matrix are combined to form a heterogeneous network. The graph neural network (GCN) is used to extract features from the heterogeneous network to obtain drug network features and target network features.

[0015] The drug structural features and the drug network features are fused using drug multimodal features;

[0016] The target structural features and the target network features are fused into multimodal target features.

[0017] Interaction information between drug and target is extracted from the fused drug multimodal features and the fused target multimodal features;

[0018] Based on the extracted interaction information between the drug and the target, updated features of the drug and the target are obtained, and the probability of interaction between the drug and the target is predicted by a classifier.

[0019] Furthermore, drug fingerprint similarity is calculated, including using the Tanimoto coefficient to calculate drug fingerprint similarity.

[0020] Furthermore, the target sequence similarity is calculated, including using the Smith-Waterman algorithm to calculate the target sequence similarity.

[0021] Furthermore, the target structure map is obtained from the Fasta sequence of the target, including: processing the Fasta sequence of the target using a protein language model to obtain the target structure map.

[0022] Furthermore, drug structure diagrams and drug fingerprints are obtained from the SMILES sequences of the drug, including: processing the SMILES sequences of the drug using the Rdkit tool to obtain drug structure diagrams and drug fingerprints.

[0023] Furthermore, the drug fingerprint similarity and the drug Gaussian similarity are fused to obtain the comprehensive drug similarity, including:

[0024] Normalize drug fingerprint similarity and drug Gaussian similarity;

[0025] The K-nearest neighbor algorithm is used to evaluate the local affinity between drugs in the drug fingerprint similarity matrix and the drug Gaussian similarity matrix, and the top 10% of local affinity is selected, while the rest are set to zero; this process is iterated until a specific stopping condition is met.

[0026] After the iteration is completed, the drug fingerprint similarity matrix and the drug Gaussian similarity matrix are added together and their average value is calculated.

[0027] Add the mean matrix to its transpose matrix, and then calculate the mean again to obtain the comprehensive drug similarity matrix;

[0028] The target sequence similarity and the target Gaussian similarity are fused to obtain the comprehensive target similarity, including:

[0029] Normalize the target sequence similarity and the target Gaussian similarity;

[0030] The K-nearest neighbor algorithm is used to evaluate the local affinity between target points in the target sequence similarity matrix and the target Gaussian similarity matrix, and the top 10% of local affinity is selected, while the rest are set to zero; this process is iterated until a specific stopping condition is met.

[0031] After the iteration is completed, the target sequence similarity matrix and the target Gaussian similarity matrix are added together and their average value is calculated.

[0032] Add the mean matrix to its transpose matrix and calculate the mean matrix again to obtain the target comprehensive similarity matrix.

[0033] Furthermore, the drug structure diagram includes an atom-to-atom diagram and a bond-to-bond diagram. In the atom-to-atom diagram, atoms in the drug are considered as nodes and covalent bonds as edges. In the bond-to-bond diagram, covalent bonds are considered as nodes and the angles between covalent bonds are considered as edges. The drug structure features are obtained from the drug structure diagram using a geometry-enhanced graph neural network (GeoGNN), including: using GeoGNN to extract drug structure features from the atom-to-atom diagram and the bond-to-bond diagram, thereby incorporating the atom-bond-angle geometric relationships of the drug.

[0034] Furthermore, the target structure map uses amino acid residues as nodes and residue contact maps as edges. A geometric encoder is used to obtain target structure features from the target structure map, including:

[0035] GCN is used to extract residue features from the target structure map. GCN only considers sparse interactions and ignores the global interactions between all residue pairs. The representation of each residue is updated through a self-attention mechanism. Attention weights are calculated using the query matrix and the bond matrix. The value matrix is ​​then weighted and summed to obtain the updated residue representation.

[0036] To incorporate the geometric information of the target, a local framework is constructed for each residue using carbon, nitrogen, and α-carbon.

[0037] v i1 =x N(i) -x Cα(i) ;

[0038] v i2 =x C(i) -x Cα(i) ;

[0039] v i1 and v i2 It is a vector determined by the coordinates of the nitrogen (N), carbon (C), and α-carbon (Cα) atoms of residue i;

[0040] u i1 ,u i2 =GramSchmidt(v i1 ,v i2 );

[0041] The GramSchmidt procedure is used to process vector v i1 and v i2 Orthogonalize them to ensure they are orthogonal to each other in space;

[0042]

[0043] R i It is the local framework of residue i, composed of orthogonal vector u i1 and u i2 And their cross products constitute these vectors, which are then normalized to form unit vectors;

[0044]

[0045] p i It is the spatial representation of residue i, which is calculated by weighted summation of information from all other residues j;

[0046] When the entire structure rotates, the local frame R... i and relative position x Cα(i) -x Cβ(j) It will also rotate accordingly, thus keeping the calculation result unchanged.

[0047] The updated residue representation and spatial representation are fused to obtain the final residue attention representation.

[0048] Furthermore, the drug structural features and the drug network features are fused using drug multimodal features, including:

[0049] The drug network features and drug structural features are concatenated, and three linear transformations are applied to obtain the query matrix, key matrix, and value matrix. The query matrix and key matrix are used to calculate the attention weights, including the correlations between network features, between network features and structural features, between structural features and network features, and between structural features. The attention weights are used to weight the elements in the value matrix to obtain the fusion features of the drug.

[0050] The target structural features and the target network features are fused into multimodal target features, including:

[0051] The target network features and target structural features are concatenated, and three linear transformations are applied to obtain the query matrix, key matrix, and value matrix. The query matrix and key matrix are used to calculate the attention weights, including the correlation between network features, the correlation between network features and structural features, the correlation between structural features and network features, and the correlation between structural features and structural features. The attention weights are used to weight the elements in the value matrix to obtain the fused features of the target.

[0052] Furthermore, interaction information between the drug and the target is extracted from the fused drug multimodal features and the fused target multimodal features, including:

[0053] For drugs, the query matrix is ​​obtained using drug features, the key matrix and value matrix are obtained using target features, the attention weights are obtained through the query matrix and key matrix, and finally, the attention weights are used to weight the elements in the value matrix, thereby updating the drug features with the help of target features.

[0054] For the target, the query matrix is ​​obtained using target features, and the key and value matrices are obtained using drug features. Attention weights are obtained through the query and key matrices. Finally, the attention weights are used to weight the elements in the value matrix, thereby updating the target features with the help of drug features.

[0055] The advantages and positive effects of this invention are as follows: Compared with common deep learning methods, this invention can not only extract complex 3D geometric information, but also uses an efficient multimodal feature fusion method to fuse structural and network features. Furthermore, it extracts interaction information between drugs and targets, and finally uses a multilayer perceptron (MLP) to predict drug-target interactions. This invention helps improve the accuracy of drug-target interaction prediction, providing valuable insights for actual disease diagnosis and treatment, as well as subsequent drug development. Attached Figure Description

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

[0057] Figure 1 This is a flowchart of the drug target interaction prediction method based on geometric characterization in an embodiment of the present invention;

[0058] Figure 2 This is a flowchart illustrating the heterogeneous network construction process in an embodiment of the present invention.

[0059] Figure 3 This is a diagram of the GeoGNN model architecture in an embodiment of the present invention;

[0060] Figure 4 This is a diagram of the multimodal feature fusion module in an embodiment of the present invention;

[0061] Figure 5 This is a diagram of the interactive information extraction module in an embodiment of the present invention. Detailed Implementation

[0062] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0063] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0064] like Figure 1 As shown in the figure, a drug target interaction prediction method based on geometric representation in this invention mainly includes three parts: data preparation, model building, and model testing.

[0065] (I) Data Preparation:

[0066] For drugs, the SMILES sequences (Simplified Molecular Input Line Entry System, an open-source cheminformatics and machine learning software suite written in C++ and Python and designed to process and analyze chemical data) are processed using the Rdkit tool to obtain drug structure maps and drug fingerprint MACCS keys (Molecular AccessSystem keys, a widely used two-dimensional structural fingerprint, mainly used in cheminformatics research fields such as molecular similarity comparison, molecular classification, molecular clustering, and molecular screening). Drug fingerprint similarity is calculated using the Tanimoto coefficient (also known as the generalized Jaccard similarity coefficient, an index used to compare the similarity between two sets (usually molecules in cheminformatics)).

[0067] For each target, a protein language model (ESM, a method using deep learning to predict protein structure and function) is used to process its Fasta sequence (a text format widely used to represent nucleic acid (DNA or RNA) and protein sequences) to obtain a target structure map. The Smith-Waterman algorithm (a dynamic programming algorithm for sequence alignment) is then used to calculate target sequence similarity. Drug Gaussian similarity and target Gaussian similarity are obtained through the drug-target association matrix.

[0068] (II) Model Building:

[0069] Step 1: Similarity fusion.

[0070] This method combines two similarity metrics for drugs and targets. Taking drugs as an example, fingerprint similarity and Gaussian similarity are normalized. Then, the K-Nearest Neighbors (KNN) algorithm is used to evaluate the local affinity between drugs in these two similarity matrices, selecting the top 10% of local affinity and setting the rest to zero. This process iterates until a specific stopping condition is met. After iteration, the two similarity matrices are summed and their average is calculated. Finally, to obtain the comprehensive similarity matrix for drugs, this average matrix is ​​added to its transpose, and the average is calculated again. The same method is used to calculate the comprehensive similarity matrix for targets.

[0071] Step 2: Network feature extraction.

[0072] A heterogeneous network of drug targets is constructed based on comprehensive drug similarity, comprehensive target similarity, and drug-target correlation matrix, such as... Figure 2 As shown, GCN is used to learn drug network features and target network features from heterogeneous networks.

[0073] Step 3: Structural feature extraction.

[0074] For drugs, drug structure diagrams include atom-to-atom diagrams and bond-to-bond diagrams. In an atom-to-atom diagram, atoms in the drug are considered nodes and covalent bonds are considered edges. In a bond-to-bond diagram, covalent bonds are considered nodes and the angles between covalent bonds are considered edges. GeoGNN is used to extract drug structural features from these two types of diagrams, thus incorporating the atom-bond-angle geometric relationships of the drug. The GeoGNN workflow is as follows: Figure 3 As shown.

[0075] For the target, the target structure graph uses amino acid residues as nodes and residue contact graphs as edges. GCN is used to extract residue features from the target structure graph. Due to computational complexity limitations, GCN only considers sparse interactions and ignores global interactions between all residue pairs. The representation of each residue is updated through a self-attention mechanism. Attention weights are calculated using the query matrix and bond matrix, and the value matrix is ​​weighted and summed using the weights to obtain the updated residue representation.

[0076] To incorporate the geometric information of the target, a local framework is constructed for each residue using carbon, nitrogen, and α-carbon.

[0077] v i1 =x N(i) -x Cα(i) ;

[0078] vi2 =x C(i) -x Cα(i) ;

[0079] v i1 and v i2 It is a vector determined by the coordinates of the nitrogen (N), carbon (C), and α-carbon (Cα) atoms of residue i.

[0080] u i1 ,u i2 =GramSchmidt(v i1 ,v i2 );

[0081] The Gram-Schmidt process (a Gram-Schmidt orthogonalization process in linear algebra that transforms a set of linearly independent vectors into an orthogonal or orthonormal basis) is used to transform vector v i1 and v i2 Orthogonalize them to ensure that they are orthogonal to each other in space.

[0082]

[0083] R i It is the local framework of residue i, composed of orthogonal vector u i1 and u i2 And their cross products constitute these vectors, which are then normalized to form unit vectors.

[0084]

[0085] p i It is the spatial representation of residue i, which is calculated by weighted summation of information from all other residues j.

[0086] When the entire structure rotates, the local frame R... i and relative position x Cα(i) -x Cβ(j) The structure will also rotate, thus maintaining the calculation results. This ensures that regardless of the structure's rotation, the model can correctly handle the relative spatial relationships between amino acid residues, unaffected by the rotation.

[0087] The updated residue representation and the spatial representation are fused to obtain the final residue attention representation. This representation considers not only the characteristics of the residues themselves, but also their relative positions in space, enhancing the model's sensitivity to geometry.

[0088] Step 4: Multimodal feature fusion.

[0089] The drug network features and drug structural features are concatenated, and three linear transformations are applied to obtain a query matrix, a key matrix, and a value matrix. Attention weights are calculated using the query matrix and the key matrix, including correlations between network features, between network features and structural features, between structural features and network features, and between structural features. Finally, the attention weights are used to weight the elements in the value matrix to obtain the fused features of the drug. The fused features of the target are obtained in the same way. This fusion method considers both intra-modal and inter-modal correlations. The multimodal feature fusion process is as follows: Figure 4 As shown.

[0090] Step 5: Extracting interactive information.

[0091] To effectively extract interaction information, an interaction information extraction module is designed to estimate the contribution of the target to the drug or the drug to the target. Specifically, for the drug, a query matrix is ​​obtained using drug features, and a key matrix and value matrix are obtained using target features. Attention weights are then obtained from the query matrix and key matrix. Finally, the attention weights are used to weight the elements in the value matrix, thereby updating the drug features with the help of the target features. For the target, a query matrix is ​​obtained using target features, and a key matrix and value matrix are obtained using drug features. Attention weights are then obtained from the query matrix and key matrix. Finally, the attention weights are used to weight the elements in the value matrix, thereby updating the target features with the help of the drug features. The interaction information extraction process is as follows: Figure 5 As shown.

[0092] (III) Model Testing:

[0093] Step 1: Use the Rdkit tool to obtain the drug structure map and drug fingerprint from the drug's SMILES sequence. Use the protein language model (ESM) to obtain the target structure map from the target's Fasta sequence. Calculate the drug fingerprint similarity using the Tanimoto coefficient. Apply the Smith-Waterman algorithm to calculate the target sequence similarity. Obtain the Gaussian similarity between the drug and the target through the drug-target association matrix. Apply similarity fusion to the two similarities of the drug and the target to obtain the comprehensive similarity of the drug and the target. Use GeoGNN to obtain the drug structure features from the drug structure map. Use a geometric encoder to obtain the target structure features from the target structure map. Combine the comprehensive similarity of the drug and the target with the drug-target association matrix to form a heterogeneous network. Then, use GCN to obtain the network features of the drug and the target. Fuse the structural features and network features of the drug and the target. Next, extract the interaction information between the drug and the target. Finally, use a multilayer perceptron (MLP) to predict the probability of interaction between the drug and the target.

[0094] Step 2: Test various parameters and key modules that affect model performance. Observe the impact of structural feature extraction, network feature extraction, multimodal feature fusion, and interaction information extraction modules on model performance. Develop model variants that remove structural feature extraction, network feature extraction, multimodal feature extraction, and interaction information extraction modules, and test the impact of different modules on model performance.

[0095] Compared to common deep learning methods, the prediction method in the above embodiments can not only extract complex 3D geometric information, but also uses an efficient multimodal feature fusion method to fuse structural and network features. Furthermore, it extracts interaction information between the drug and the target, and finally uses a multilayer perceptron (MLP) to predict drug-target interactions. This invention helps improve the accuracy of drug-target interaction prediction, providing valuable insights for actual disease diagnosis and treatment, as well as subsequent drug development.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A drug target interaction method based on geometric characterization, characterized in that, The method includes: Drug structure maps and drug fingerprints are obtained from the SMILES sequences of the drug. The drug structure maps include atom-to-atom maps and bond-to-bond maps. In the atom-to-atom map, atoms in the drug are considered as nodes and covalent bonds as edges. In the bond-to-bond map, covalent bonds are considered as nodes and the angles between covalent bonds are considered as edges. The drug structure features are obtained from the drug structure maps using a geometry-enhanced graph neural network (GeoGNN), including: extracting drug structure features from the atom-to-atom map and bond-to-bond map using GeoGNN, thereby incorporating the atom-bond-angle geometric relationships of the drug. Obtain the target structure map from the Fasta sequence of the target; use a geometric encoder to obtain the target structure features from the target structure map; Calculate the drug fingerprint similarity of the drug and the target sequence similarity of the target separately; The Gaussian similarity of drugs and targets is obtained through the drug-target correlation matrix; The drug fingerprint similarity and the drug Gaussian similarity are fused to obtain the comprehensive drug similarity, including: normalizing the drug fingerprint similarity and the drug Gaussian similarity; The K-nearest neighbor algorithm is used to evaluate the local affinity between drugs in the drug fingerprint similarity matrix and the drug Gaussian similarity matrix, and the top 10% of local affinity is selected, while the rest are set to zero; this process is iterated until a specific stopping condition is met. After the iteration is completed, the drug fingerprint similarity matrix and the drug Gaussian similarity matrix are added together and their average value is calculated. Add the mean matrix to its transpose matrix, and then calculate the mean again to obtain the comprehensive drug similarity matrix; The target sequence similarity and the target Gaussian similarity are fused to obtain the comprehensive target similarity, including: Normalize the target sequence similarity and the target Gaussian similarity; The K-nearest neighbor algorithm is used to evaluate the local affinity between target points in the target sequence similarity matrix and the target Gaussian similarity matrix, and the top 10% of local affinity is selected, while the rest are set to zero; this process is iterated until a specific stopping condition is met. After the iteration is completed, the target sequence similarity matrix and the target Gaussian similarity matrix are added together and their average value is calculated. Add the mean matrix to its transpose matrix, and then calculate the mean again to obtain the target comprehensive similarity matrix; The target sequence similarity and the target Gaussian similarity are fused to obtain the comprehensive target similarity. The comprehensive similarity of the drugs, the comprehensive similarity of the targets, and the drug-target correlation matrix are combined to form a heterogeneous network. The graph neural network (GCN) is used to extract features from the heterogeneous network to obtain drug network features and target network features. The drug structural features and drug network features are fused using a multimodal feature fusion method, which includes: concatenating the drug network features and drug structural features, applying three linear transformations to obtain a query matrix, a key matrix, and a value matrix, and using the query matrix and key matrix to calculate attention weights, including the correlations between network features, between network features and structural features, between structural features and network features, and between structural features; the attention weights are used to weight the elements in the value matrix to obtain the fused features of the drug. The target structural features and target network features are fused using a multimodal feature fusion method, which includes: concatenating the target network features and target structural features, applying three linear transformations to obtain a query matrix, a key matrix, and a value matrix, and using the query matrix and key matrix to calculate attention weights, including the correlations between network features, between network features and structural features, between structural features and network features, and between structural features; the attention weights are used to weight the elements in the value matrix to obtain the fused features of the target. Interaction information between the drug and target is extracted from the fused drug multimodal features and the fused target multimodal features, including: For drugs, the query matrix is ​​obtained using drug features, the key matrix and value matrix are obtained using target features, the attention weights are obtained through the query matrix and key matrix, and finally, the attention weights are used to weight the elements in the value matrix, thereby updating the drug features with the help of target features. For the target, the query matrix is ​​obtained using the target features, the key matrix and value matrix are obtained using the drug features, the attention weights are obtained through the query matrix and key matrix, and finally, the attention weights are used to weight the elements in the value matrix, thereby updating the target features with the help of the drug features. Based on the extracted interaction information between the drug and the target, updated features of the drug and the target are obtained, and the probability of interaction between the drug and the target is predicted by a classifier.

2. The drug target interaction method based on geometric characterization according to claim 1, characterized in that, Calculating drug fingerprint similarity includes: using the Tanimoto coefficient to calculate drug fingerprint similarity.

3. The drug target interaction method based on geometric characterization according to claim 1, characterized in that, Calculate the target sequence similarity, including using the Smith-Waterman algorithm to calculate the target sequence similarity.

4. The drug target interaction method based on geometric characterization according to claim 1, characterized in that, Obtaining target structure maps from the Fasta sequence of the target site includes: processing the Fasta sequence of the target site using a protein language model to obtain the target structure map.

5. The drug target interaction method based on geometric characterization according to claim 1, characterized in that, Obtaining drug structure maps and drug fingerprints from drug SMILES sequences includes: processing drug SMILES sequences using the Rdkit tool to obtain drug structure maps and drug fingerprints.

6. The drug target interaction method based on geometric characterization according to claim 1, characterized in that, The target structure map uses amino acid residues as nodes and residue contact maps as edges. A geometric encoder is used to extract target structure features from the target structure map, including: GCN is used to extract residue features from the target structure map. GCN only considers sparse interactions and ignores the global interactions between all residue pairs. The representation of each residue is updated through a self-attention mechanism. Attention weights are calculated using the query matrix and the bond matrix. The value matrix is ​​then weighted and summed to obtain the updated residue representation. To incorporate the geometric information of the target, a local framework is constructed for each residue using carbon, nitrogen, and α-carbon. ； ； and It is composed of residues i The vector determined by the coordinates of nitrogen (N), carbon (C), and α-carbon (Cα) atoms; , , Residues i The coordinates of nitrogen (N), α-carbon (Cα), and carbon (C) atoms; ； The GramSchmidt procedure is used to process vectors and Orthogonalize them to ensure they are orthogonal to each other in space; ； It is a residue i The local framework is composed of orthogonal vectors and And their cross products constitute these vectors, which are then normalized to form unit vectors; = ； It is a residue i The spatial representation, which is achieved by using all other residues j The information is weighted and summed to calculate; residues i Other residues j Information weight, Residues j of carbon The coordinates of the atom; When the entire structure rotates, the local frame... and relative position It will also rotate accordingly, thus keeping the calculation result unchanged; The updated residue representation and spatial representation are fused to obtain the final residue attention representation.

Images