Drug design method, device, equipment and medium based on topology data analysis

Abstract

The application discloses a drug design method and device based on topological data analysis, equipment and medium, relates to the technical field of computer-aided drug design, and the method comprises the steps of obtaining a plurality of three-dimensional structure conformations of a target protein, determining the binding pocket region of each three-dimensional structure conformation, and constructing a conformation set; calculating a topological descriptor representing the topological relationship between atoms based on the atomic coordinates of the binding pocket region of each three-dimensional structure conformation, and converting the topological descriptor into a corresponding conformation feature vector; performing a pooling operation on the conformation feature vector corresponding to each three-dimensional structure conformation to generate a unified topological feature representation representing the common topological feature mode of the target protein binding pocket under a plurality of three-dimensional structure conformations; and screening or generating candidate drugs based on the unified topological feature representation. The common topological invariant information of a plurality of conformations is used to represent the flexible protein binding pocket, the screening omission caused by single conformation representation is reduced, and the representation accuracy and the calculation efficiency are considered.

Description

Technical Field

[0001] This invention relates to the field of computer-aided drug design technology, and in particular to drug design methods, apparatus, equipment and media based on topology data analysis. Background Technology

[0002] In antiviral drug development, computer-aided drug design can significantly shorten the development cycle and reduce costs. Structure-based drug design, in particular, relies on the three-dimensional structure of target proteins (such as neuraminidase or hemagglutinin in influenza viruses). However, a long-standing technical challenge is that proteins are not rigid structures in physiological states, but rather undergo various conformational changes, such as side-chain rotation, loop movement, and relative movement of domains. Traditional methods, such as molecular docking, typically treat proteins as rigid or allow only limited flexibility, which leads to: It is impossible to accurately predict the induced fit effect between ligands and proteins during the dynamic binding process.

[0003] It is possible to miss highly active compounds that bind to non-preferred conformations.

[0004] Traditional methods are less robust for targets like influenza viruses, which are prone to altering their local conformation through mutation.

[0005] Current attempts to sample multiple conformations through molecular dynamics simulations or use integrated docking are extremely costly and difficult to apply to large-scale virtual screening. Therefore, there is an urgent need for a new method that can effectively handle protein flexibility and has high computational efficiency. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a drug design method, apparatus, device, and medium based on topological data analysis, which can achieve stable characterization of flexible target protein binding pockets while taking into account computational efficiency, and thereby conduct candidate drug screening or generation. The specific scheme is as follows: In a first aspect, this application discloses a drug design method based on topological data analysis, comprising: Multiple three-dimensional structural conformations of the target protein are obtained, and the binding pocket region of each three-dimensional structural conformation is determined to construct a conformation set. Based on the atomic coordinates of the combined pocket region of each of the three-dimensional structural conformations, a topological descriptor characterizing the topological relationship between atoms is calculated, and the topological descriptor is converted into the corresponding conformation feature vector. Pooling operations are performed on the conformation feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations. Candidate drugs can be screened or generated based on the unified topological feature representation.

[0007] Optionally, the calculation of the topological descriptor characterizing the topological relationships between atoms based on the atomic coordinates of the combined pocket regions of each of the three-dimensional structural conformations includes: The target point cloud is constructed based on the atomic coordinates of the pocket region combined with the three-dimensional structural conformation. When the current filter value is set to the preset minimum filter value, each atom in the target point cloud is a connected component. Record the occurrence filter value of each connected component during the current filter value increase process and the disappearance filter value when it first merges and disappears, as 0-dimensional topological features; Record the appearance filter value of each annular or porous structure during the current filter value increase process and the disappearance filter value when it is first filled and disappears, as a 1D topological feature; A topological descriptor for the three-dimensional structural conformation is generated based on the 0-dimensional topological features and the 1-dimensional topological features; wherein the topological descriptor is any one or more of persistent graphs, persistent barcodes, or persistent Laplacian spectral features.

[0008] Optionally, performing pooling operations on the conformational feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation characterizing the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations includes: The weight parameters corresponding to each conformation feature vector are determined based on the energy stability parameter of the conformation feature vector corresponding to each of the three-dimensional structural conformations and the similarity parameter between the conformation feature vector and other conformation feature vectors in the conformation set. Dynamic pooling is performed on each of the conformational feature vectors based on the weight parameters to generate a unified topological feature representation that characterizes the common topological feature patterns of the target protein binding pocket under multiple three-dimensional structural conformations.

[0009] Optionally, determining the weight parameters corresponding to each conformation feature vector based on the energy stability parameter of the conformation feature vector corresponding to each of the three-dimensional structural conformations and the similarity parameter between the conformation feature vector and other conformation feature vectors in the conformation set includes: Obtain the energy stability parameters corresponding to each of the three-dimensional structural conformations, and calculate the similarity matrix between the feature vectors of each conformation; wherein each element in the similarity matrix represents the similarity between any two feature vectors of the conformations; The average similarity between each feature vector and other feature vectors of the same configuration is calculated based on the similarity matrix. The energy stability parameters are normalized to obtain normalized energy stability parameters. Based on the average similarity and the normalized energy stability parameter, a corresponding weight parameter is assigned to each conformation feature vector. Accordingly, the step of performing dynamic pooling operations on each conformational feature vector based on each of the weight parameters to generate a unified topological feature representation characterizing the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations includes: The weight parameters are used as attention scores for the corresponding conformational feature vectors. The conformational feature vectors are then weighted and summed based on the attention scores to generate a unified topological feature representation that characterizes the common topological feature patterns of the target protein binding pocket under multiple three-dimensional structural conformations.

[0010] Optionally, after generating a unified topological feature representation of the common topological feature patterns of the target protein binding pocket under multiple three-dimensional structural conformations, the method further includes: The amino acid sequence of the target protein is obtained, and the sequence embedding vector of the amino acid sequence is extracted by a pre-trained protein language model. The unified topological feature representation and the sequence embedding vector are fused using a cross-attention mechanism to generate a multimodal joint feature representation; Accordingly, the screening or generation of candidate drugs based on the unified topological feature representation includes: Candidate drugs can be screened or generated based on the multimodal joint feature representation.

[0011] Optionally, screening candidate drugs based on the unified topological feature representation includes: Calculate the molecular topological feature representation of each compound in a predefined set of compounds; Calculate the similarity information between the molecular topological feature representations of each compound and the unified topological feature representation; If the similarity information is greater than or equal to a preset similarity threshold, the compound is identified as a candidate drug.

[0012] Optionally, before generating candidate drugs based on the unified topological feature representation, the method further includes: Historical unified topological feature representations are input into the initial molecular generation model to generate historical drug candidates; The matching degree information between the molecular topological feature representation of the historical candidate drug and the historical unified topological feature representation is used as the reward signal; By using reinforcement learning to update the parameters of the initial molecular generation model based on the reward signal, and iteratively training until a preset convergence condition is met, a molecular generation model for generating candidate drugs is obtained.

[0013] Secondly, this application discloses a drug design device based on topological data analysis, comprising: The conformation acquisition module is used to acquire multiple three-dimensional structural conformations of the target protein and determine the binding pocket region of each three-dimensional structural conformation to construct a conformation set. The descriptor generation module is used to calculate a topological descriptor representing the topological relationship between atoms based on the atomic coordinates of the combined pocket region of each of the three-dimensional structural conformations, and to convert the topological descriptor into the corresponding conformation feature vector. The pooling module is used to perform pooling operations on the conformation feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations. A drug design module is used to screen or generate candidate drugs based on the unified topological feature representation.

[0014] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the aforementioned disclosed drug design method based on topology data analysis.

[0015] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed drug design method based on topology data analysis.

[0016] As can be seen, this application obtains multiple three-dimensional structural conformations of the target protein and performs topological data analysis on the binding pocket regions of each conformation to obtain the topological descriptors and conformational feature vectors corresponding to each conformation. Then, the conformational feature vectors of multiple conformations are pooled and fused to form a unified topological feature representation capable of characterizing common topological feature patterns. Compared to schemes that characterize based on only a single conformation, this application can reduce the impact of local coordinate perturbations on the characterization results while preserving the overall spatial shape and connectivity information of the binding pocket, and is beneficial in reducing the undersight of non-dominant but functionally relevant conformations. Compared to schemes that independently perform complex docking calculations on multiple conformations, this application uses a unified feature representation to participate in candidate drug screening or generation, which helps reduce computational overhead and improves applicability in large-scale virtual screening scenarios. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a flowchart of a drug design method based on topology data analysis disclosed in this application; Figure 2 These are two different conformations of the same influenza virus neuraminidase protein disclosed in this application; Figure 3 Example conformational diagrams obtained by topological data analysis of the active pocket regions of two different conformations of the same influenza virus neuraminidase protein disclosed in this application. Figure 4 This is a schematic diagram of a drug design device based on topology data analysis disclosed in this application; Figure 5 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In antiviral drug development, computer-aided drug design can significantly shorten the development cycle and reduce costs. Structure-based drug design, in particular, relies on the three-dimensional structure of target proteins (such as neuraminidase or hemagglutinin in influenza viruses). However, a long-standing technical challenge is that proteins are not rigid structures in physiological states, but rather undergo various conformational changes, such as side-chain rotation, loop movement, and relative movement of domains. Traditional methods, such as molecular docking, typically treat proteins as rigid or allow only limited flexibility, which leads to: It is impossible to accurately predict the induced fit effect between ligands and proteins during the dynamic binding process.

[0021] It is possible to miss highly active compounds that bind to non-preferred conformations.

[0022] Traditional methods are less robust for targets like influenza viruses, which are prone to altering their local conformation through mutation.

[0023] Current attempts to sample multiple conformations through molecular dynamics simulations or use integrated docking are extremely costly and difficult to apply to large-scale virtual screening. Therefore, there is an urgent need for a new method that can effectively handle protein flexibility and has high computational efficiency.

[0024] Therefore, this invention provides a drug design scheme based on topological data analysis, which can accurately utilize the topological data of target proteins in drug design to achieve rapid and accurate drug design.

[0025] like Figure 1 As shown, this invention provides a drug design method based on topological data analysis, comprising: Step S11: Obtain multiple three-dimensional structural conformations of the target protein and determine the binding pocket region of each three-dimensional structural conformation to construct a conformation set.

[0026] In this embodiment, multiple three-dimensional structural conformations of the target protein under different physiological states are obtained, specifically through the following methods: In one specific implementation, a whole-atom molecular dynamics simulation is performed on the target protein. The protein's motion trajectory is sampled at a preset time scale, multiple conformation frames are extracted at preset time intervals, and redundant information is removed by cluster analysis to obtain a conformation set covering the main conformation subset of the protein. The preset time scale is nanoseconds or microseconds.

[0027] In another specific implementation, the crystal structure of the target protein in the binding state of different small molecules (inhibitors, substrate analogs) is obtained from a protein database or by means of X-ray crystallography, cryo-electron microscopy, etc. The crystal structure represents the binding pocket conformation of the protein under different ligand induction, so as to obtain a set of conformations.

[0028] In another specific implementation, a reasonable conformational change of the target protein binding pocket region is generated by using computational methods based on normal mode analysis, principal component analysis, or random conformational enumeration without performing long-term kinetic simulations, thus forming a conformational set.

[0029] In the set of conformations obtained above, each three-dimensional structural conformation contains complete atomic coordinate information for subsequent calculation of topological descriptors. The number of conformations can be adjusted according to the flexibility of the protein and computational resources. In this embodiment, 5 to 20 representative conformations are selected to balance characterization accuracy and computational efficiency.

[0030] Furthermore, for each obtained three-dimensional structural conformation, its binding pocket region is further determined, that is, the active cavity region on or inside the protein that can specifically bind to drug molecules is identified. The specific method for determining the binding pocket region is as follows: If the target protein already has a co-crystal structure, the atomic coordinates of all amino acid residues within a certain distance (e.g., 4.0–6.0 Å) around the ligand can be taken as the binding pocket region based on the position of the small molecule ligand.

[0031] Alternatively, binding pocket identification software can be used to identify potential binding pockets based on the geometry of depressions and cavities on the protein surface, and the most likely pockets can be selected as the final binding pocket regions based on indicators such as volume and hydrophobic surface area.

[0032] Alternatively, multiple sequence alignment of the protein amino acid sequence can be performed to calculate the conservation score of each site. Regions with high conservation scores usually correspond to functionally important binding sites, which helps to determine the pocket boundary and thus identify the binding pocket region based on the pocket boundary.

[0033] After identifying the binding pocket region, the three-dimensional coordinates of all non-hydrogen atoms within this region are extracted for subsequent topological descriptor calculations. This approach, analyzing only the pocket region, eliminates irrelevant structural noise far from the binding interface, reduces the point cloud size and computational complexity, and allows the extracted topological features to be more focused on functional regions related to drug binding.

[0034] like Figure 2 The diagram shows two different conformations of the same influenza virus neuraminidase protein, namely conformation A and conformation B. Figure 2 It can be seen that the atomic positions of the two conformations are different, but there are common feature points in their conformation diagrams, which can be used to extract common topological features later.

[0035] Step S12: Calculate a topological descriptor representing the topological relationship between atoms based on the atomic coordinates of the combined pocket region of each of the three-dimensional structural conformations, and convert the topological descriptor into the corresponding conformation feature vector.

[0036] In this embodiment, a target point cloud is constructed based on the atomic coordinates of the pocket region of the three-dimensional structural conformation. When the current filter value is set to a preset minimum filter value, each atom in the target point cloud is considered a connected component. The appearance filter value and the disappearance filter value when each connected component first merges and disappears during the increase of the current filter value are recorded as 0-dimensional topological features. The appearance filter value and the disappearance filter value when each ring or hole structure first disappears after being filled during the increase of the current filter value are recorded as 1-dimensional topological features. A topological descriptor for the three-dimensional structural conformation is generated based on the 0-dimensional and 1-dimensional topological features. The topological descriptor is any one or more of persistent graphs, persistent barcodes, or persistent Laplacian spectra. It is understood that, taking a single protein conformation as an example, starting from atomic coordinates, the birth and death events of topological features are recorded through continuous changes in the filter value, ultimately generating a topological descriptor for that conformation. This embodiment is applicable to each conformation in the conformation set.

[0037] Specifically, a target point cloud is constructed based on the three-dimensional coordinates of all non-hydrogen atoms in the binding pocket region within the current protein conformation. ,in, For the first The coordinates of each atom are used to define a filtering function. For example, if the filtering function uses the Euclidean distance from the atom to the geometric center of the bounding pocket: Where c represents the coordinates of the pocket center point, and the value of the filtering function is the filter value. The current filter value is set to the preset minimum value. This value is less than the minimum distance between any two points in the point cloud. At this filter value, each atom in the point cloud is an isolated connected component (0-dimensional simplex), without any edges, loops, or holes; then, the filter value is gradually increased while tracking the changes in the connected components in tiny steps. The filter value is gradually increased. Whenever the filter value exceeds the distance between two points, an edge is formed between them. When the filter value exceeds the radius of the circumcircle of the triangle, the triangle is filled. During this process, connected components appear and merge. Specifically, when the filter value first makes an atom the core of a local minimum region, the appearance filter value of that connected component is recorded. When the filter value increases to the point where a connection path appears between two connected components, the two components merge into one, and the disappearance filter value of the merged component (i.e., the filter value at the time of merging) is recorded. All appearance-disappearance filter value pairs of connected components constitute a 0-dimensional topological feature set. As the filter value increases, when the connection between atoms forms a closed loop (i.e., a 1-dimensional cycle) and the loop has not yet been filled by the triangle, the appearance filter value of the loop structure is recorded. As the filter value further increases, when the internal region of the loop is completely filled (i.e., the loop becomes the boundary of a higher-dimensional simplex), the disappearance filter value of the loop structure is recorded. All appearance-disappearance filter value pairs of loop or hole structures constitute a 1-dimensional topological feature set. Each occurrence-disappearance filter pair in the above 0-dimensional and 1-dimensional topological features is mapped to a point in a two-dimensional plane, where the horizontal axis represents the occurrence filter value and the vertical axis represents the disappearance filter value, resulting in a persistent graph; or each feature is represented as a line segment spanning [occurrence filter value, disappearance filter value], resulting in a persistent barcode; or further, the combined Laplacian matrix in the persistent homology process is subjected to spectral decomposition to extract the curve of feature value change with filter value, resulting in persistent Laplacian spectral features. Therefore, the final output topological descriptor can be any one or a combination of persistent graphs, persistent barcodes, or persistent Laplacian spectral features.

[0038] For each topological descriptor (e.g., persistent image) generated from a configuration, it needs to be converted into a fixed-length numerical vector for subsequent pooling fusion. This embodiment uses a persistent image method for conversion, with the following specific steps: Discretize the two-dimensional plane of the persistent image (horizontal axis represents the appearance filter value, and vertical axis represents the disappearance filter value) into an N×N uniform grid. For each persistent point (i.e., each topological feature), place a two-dimensional Gaussian kernel function centered on its coordinates. Accumulate the contribution values ​​of all Gaussian kernel functions at each grid point to obtain the pixel value of that grid point; flatten the N×N grid into a length of N... 2The numerical vector is the configuration feature vector of the configuration. In addition to persistent images, persistent spectra, persistent entropy or other kernel methods can also be used to convert persistent images into vectors, which will not be elaborated further.

[0039] Furthermore, in addition to the aforementioned continuous cohomology, Morse theory can be used for topological data analysis. Specifically, by applying Morse theory and Morse complexes to filtering functions (distance functions, electrostatic potential functions) defined on protein point clouds, the critical points (maximums, minima, saddle points) of the functions can be identified. These critical points and their connections constitute the Morse complex, which characterizes the topological skeleton of the original shape in a simplified combinatorial structure. Then, information on the variation of the topological invariants of the Morse complex with the filtering scale is extracted to generate Morse barcodes, similar to continuous barcodes, used to describe the appearance and disappearance of protein pocket connectivity, loops, holes, and other features.

[0040] Furthermore, topological data analysis can be performed using Reeb plots and the Mapper algorithm. For protein conformations, Reeb plots of their electrostatic or hydrophobic fields can be calculated. These plots visually demonstrate the topological connectivity between high- and low-value regions of potential or hydrophobicity on the protein surface. Pockets often correspond to specific topological structures. By comparing the topological structures of Reeb plots under different conformations, another type of topological descriptor can be formed.

[0041] Taking the binding pocket of influenza virus neuraminidase as an example, such as Figure 3 As shown, the 0-persistence map obtained through the above steps usually contains several high-persistence points, corresponding to the main cavity regions in the pocket; the high-persistence points in the 1-persistence map correspond to the annular channel or internal holes at the pocket entrance. These topological features are not sensitive to small displacements of atomic coordinates, but can characterize the overall shape of the pocket.

[0042] Step S13: Perform pooling operation on the conformation feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations.

[0043] In this embodiment, the weight parameters corresponding to each conformation feature vector are determined based on the energy stability parameter of the conformation feature vector corresponding to each of the three-dimensional structural conformations and the similarity parameter between the conformation feature vector and other conformation feature vectors in the conformation set. Based on the weight parameters, dynamic pooling operation is performed on each of the conformation feature vectors to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations.

[0044] Specifically, the energy stability parameters corresponding to each of the three-dimensional structural conformations are obtained, and the similarity matrix between the feature vectors of each conformation is calculated. The energy stability parameters are free energy, Boltzmann weights, or stability parameters derived from both. Each element in the similarity matrix represents the similarity between any two conformation feature vectors. Based on the similarity matrix, the average similarity between each conformation feature vector and other conformation feature vectors is calculated. The energy stability parameters are normalized to obtain normalized energy stability parameters. Based on the average similarity and the normalized energy stability parameters, corresponding weight parameters are assigned to each conformation feature vector. The higher the energy stability and the lower the average similarity, the larger the corresponding weight parameter.

[0045] Accordingly, the step of performing dynamic pooling operation on each conformation feature vector based on each of the weight parameters to generate a unified topological feature representation characterizing the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations includes: using the weight parameters as attention scores for the corresponding conformation feature vectors, and performing a weighted summation of the conformation feature vectors based on the attention scores to generate a unified topological feature representation characterizing the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations.

[0046] It is understandable that for the eigenvectors of all conformations in the conformation set... Then, where m represents the number of conformations, they are fused into a single unified topological feature representation through pooling operations. This is used to characterize the common topological feature patterns of target protein binding pockets across multiple conformations. Specifically, the pooling methods in this embodiment can be max pooling, mean pooling, attention-weighted pooling, and dynamic pooling based on conformational topological features. The specific pooling methods are as follows: Max pooling is the process of taking the maximum value of the eigenvectors of each conformation in each dimension, that is... This is done to preserve the most prominent topological response in each conformation. Mean pooling is achieved by taking the arithmetic mean of the eigenvectors of each conformation dimension by dimension, i.e. This method considers the contributions of all conformations in a balanced way. Attention-weighted pooling is used to learn trainable attention scores. Weighted summation after Softmax normalization ,in, , Si For the first i Learnable attention parameters for each conformation. Sk For the first kThe attention parameters for each conformation can be understood as follows: From the set of conformations obtained through molecular dynamics simulations, each conformation has a corresponding free energy or Boltzmann weight (energy stability parameter). Conformations with lower energy (or higher weight) are more likely to occur under physiological conditions and may be more important in binding. Therefore, in attention-weighted pooling, the network is not directly allowed to learn... Instead, The pooling process is initialized or constrained to weights proportional to the Boltzmann factor of the conformation, and then fine-tuned by a learnable scaling factor. In this way, the pooling process is a priori incorporated with physicochemical laws.

[0047] In this embodiment, dynamic pooling based on the topological uniqueness of the conformations is selected. Conformations with more unique topological features (i.e., lower average similarity to other conformations) are more important for representing the core topological pattern of the binding pocket and are assigned higher weights. The specific steps are: calculating the similarity matrix S between the feature vectors of each conformation, where... express and Similarity (cosine similarity can be used); for each conformation i The average similarity of the conformational feature vector to all other conformations is calculated, and weights are assigned based on this average similarity, with lower average similarity vectors receiving higher weights. These weights are then used as attention scores. The feature vectors of each conformation are then weighted and summed, and the sum is used as the attention score. Finally, the attention score determines the unified topological feature representation. This dynamic pooling method can highlight rare but functionally critical conformational features while suppressing interference from redundant conformations. In practical applications, any of the above pooling methods or a combination thereof can be selected based on the flexibility of the target protein and the diversity of the conformational set.

[0048] Furthermore, after obtaining the unified topological features, to further enhance the characterization ability of the protein binding pocket, a multimodal fusion method is provided. Specifically, the amino acid sequence of the target protein is obtained, and the sequence embedding vector of the amino acid sequence is extracted through a pre-trained protein language model. The unified topological feature representation and the sequence embedding vector are fused using a cross-attention mechanism to generate a multimodal joint feature representation. It can be understood that after obtaining the unified topological feature representation of the target protein, the amino acid sequence of the target protein is further obtained. The amino acid sequence is input into a pre-trained protein language model (e.g., ESM-2 or ProtBERT), and its last hidden state is extracted as the sequence embedding vector. The sequence embedding vector is fused with the unified topological feature representation using a cross-attention mechanism: the unified topological feature representation is used as the query, and the sequence embedding vector is used as the key and value. The fused multimodal joint feature representation is obtained through attention calculation. This joint feature simultaneously encodes the geometric topological information and sequence evolution information of the protein binding pocket.

[0049] Furthermore, the historical unified topological feature representation is input into the initial molecular generation model to generate historical candidate drugs. The matching degree information between the molecular topological feature representation of the historical candidate drug and the historical unified topological feature representation is used as a reward signal. The parameters of the initial molecular generation model are updated according to the reward signal through reinforcement learning, and the model is iteratively trained until a preset convergence condition is met, resulting in a molecular generation model for generating candidate drugs. It is understood that after the unified topological feature representation is determined, a scorer can first be used to evaluate the matching degree between the molecular topological feature representation of the generated molecule and the historical unified topological feature representation corresponding to the target point. The evaluation result is then fed back to the initial molecular generation model as a reward signal, enabling the trained molecular generation model to generate candidate molecules that better match the target point combined with the pocket topological pattern.

[0050] Step S14: Screen or generate candidate drugs based on the unified topological feature representation.

[0051] In this embodiment, the process of screening candidate drugs is as follows: calculating the molecular topological feature representation of each compound in the preset compound set; calculating the similarity information between the molecular topological feature representation of each compound and the unified topological feature representation; if the similarity information is greater than or equal to the preset similarity threshold, then the compound is determined as a candidate drug.

[0052] In this embodiment, the process of generating candidate drugs is as follows: inputting the unified topological feature representation or the multimodal joint feature representation into the molecular generation model to generate candidate drugs through the model.

[0053] In this embodiment, the screening or generation of candidate drugs based on the unified topological feature representation includes: screening or generating candidate drugs based on the multimodal joint feature representation. The multimodal joint feature representation replaces the unified topological feature for subsequent candidate drug screening or molecule generation. Thus, by introducing sequence information, it is beneficial to improve the sufficiency of characterization and screening stability of sequence variants or mutant targets.

[0054] As can be seen, this application performs topological data analysis on the binding pocket regions of multiple three-dimensional structural conformations of the target protein, extracts stable and discriminative topological features between each conformation, and forms a unified topological feature representation through pooling operations. Candidate drugs are then screened or generated based on this unified topological feature representation. Since the unified topological feature representation integrates the common topologically invariant information of multiple conformations, it can reduce the impact of local atomic coordinate perturbations to a certain extent and help reduce the missed screening of candidate molecules binding to non-preferred conformations. Simultaneously, using the unified feature representation in subsequent drug design processes helps reduce the computational burden of processing multiple conformations one by one, thus making it suitable for large-scale virtual screening scenarios.

[0055] like Figure 4 As shown, the present invention correspondingly discloses a drug design device based on topological data analysis, comprising: The conformation acquisition module 11 is used to acquire multiple three-dimensional structural conformations of the target protein and determine the binding pocket region of each three-dimensional structural conformation to construct a conformation set. The descriptor generation module 12 is used to calculate a topological descriptor representing the topological relationship between atoms based on the atomic coordinates of the combined pocket region of each of the three-dimensional structural conformations, and to convert the topological descriptor into the corresponding conformation feature vector. Pooling module 13 is used to perform pooling operations on the conformation feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations. Drug design module 14 is used to screen or generate candidate drugs based on the unified topological feature representation.

[0056] Therefore, by acquiring multiple protein conformations and extracting the topologically invariant features of their binding pocket regions, and using pooling fusion to generate a unified topological feature representation that is stable to conformational changes, the prediction bias caused by neglecting protein flexibility in traditional schemes can be alleviated. At the same time, the unified topological feature representation can be directly involved in subsequent candidate drug screening or generation processes, which is beneficial to control computational overhead while ensuring characterization stability. This is applicable to the development of antiviral drugs targeting rapidly mutating pathogens such as influenza virus.

[0057] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0058] Figure 5This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the drug design method based on topology data analysis disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0059] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0060] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0061] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0062] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. It can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the drug design method based on topology data analysis executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0063] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned drug design method based on topology data analysis. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0064] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0065] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0066] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0067] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A drug design method based on topological data analysis, characterized in that, include: Multiple three-dimensional structural conformations of the target protein are obtained, and the binding pocket region of each three-dimensional structural conformation is determined to construct a conformation set. Based on the atomic coordinates of the combined pocket region of each of the three-dimensional structural conformations, a topological descriptor characterizing the topological relationship between atoms is calculated, and the topological descriptor is converted into the corresponding conformation feature vector. Pooling operations are performed on the conformation feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations. Candidate drugs can be screened or generated based on the unified topological feature representation.

2. The drug design method based on topological data analysis according to claim 1, characterized in that, The calculation of topological descriptors characterizing the topological relationships between atoms based on the atomic coordinates of the combined pocket regions of each of the three-dimensional structural conformations includes: The target point cloud is constructed based on the atomic coordinates of the pocket region combined with the three-dimensional structural conformation. When the current filter value is set to the preset minimum filter value, each atom in the target point cloud is a connected component. Record the occurrence filter value of each connected component during the current filter value increase process and the disappearance filter value when it first merges and disappears, as 0-dimensional topological features; Record the appearance filter value of each annular or porous structure during the current filter value increase process and the disappearance filter value when it is first filled and disappears, as a 1D topological feature; A topological descriptor for the three-dimensional structural conformation is generated based on the 0-dimensional topological features and the 1-dimensional topological features; wherein the topological descriptor is any one or more of persistent graphs, persistent barcodes, or persistent Laplacian spectral features.

3. The drug design method based on topological data analysis according to claim 1, characterized in that, The step of performing pooling operations on the conformational feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation characterizing the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations includes: The weight parameters corresponding to each conformation feature vector are determined based on the energy stability parameter of the conformation feature vector corresponding to each of the three-dimensional structural conformations and the similarity parameter between the conformation feature vector and other conformation feature vectors in the conformation set. Dynamic pooling is performed on each of the conformational feature vectors based on the weight parameters to generate a unified topological feature representation that characterizes the common topological feature patterns of the target protein binding pocket under multiple three-dimensional structural conformations.

4. The drug design method based on topology data analysis according to claim 3, characterized in that, The step of determining the weight parameters corresponding to each conformation feature vector based on the energy stability parameter of the conformation feature vector corresponding to each of the three-dimensional structural conformations and the similarity parameter between the conformation feature vector and other conformation feature vectors in the conformation set includes: Obtain the energy stability parameters corresponding to each of the three-dimensional structural conformations, and calculate the similarity matrix between the feature vectors of each conformation; wherein each element in the similarity matrix represents the similarity between any two feature vectors of the conformations; The average similarity between each feature vector and other feature vectors of the same configuration is calculated based on the similarity matrix. The energy stability parameters are normalized to obtain normalized energy stability parameters. Based on the average similarity and the normalized energy stability parameter, a corresponding weight parameter is assigned to each conformation feature vector. Accordingly, the step of performing dynamic pooling operations on each conformational feature vector based on each of the weight parameters to generate a unified topological feature representation characterizing the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations includes: The weight parameters are used as attention scores for the corresponding conformational feature vectors. The conformational feature vectors are then weighted and summed based on the attention scores to generate a unified topological feature representation that characterizes the common topological feature patterns of the target protein binding pocket under multiple three-dimensional structural conformations.

5. The drug design method based on topological data analysis according to claim 1, characterized in that, After generating a unified topological feature representation of the common topological feature patterns of the target protein binding pocket in multiple three-dimensional structural conformations, the method further includes: The amino acid sequence of the target protein is obtained, and the sequence embedding vector of the amino acid sequence is extracted by a pre-trained protein language model. The unified topological feature representation and the sequence embedding vector are fused using a cross-attention mechanism to generate a multimodal joint feature representation; Accordingly, the screening or generation of candidate drugs based on the unified topological feature representation includes: Candidate drugs can be screened or generated based on the multimodal joint feature representation.

6. The drug design method based on topological data analysis according to claim 1, characterized in that, Screening candidate drugs based on the unified topological feature representation includes: Calculate the molecular topological feature representation of each compound in a predefined set of compounds; Calculate the similarity information between the molecular topological feature representations of each compound and the unified topological feature representation; If the similarity information is greater than or equal to a preset similarity threshold, the compound is identified as a candidate drug.

7. The drug design method based on topological data analysis according to claim 1, characterized in that, Before generating candidate drugs based on the unified topological feature representation, the following steps are also included: Historical unified topological feature representations are input into the initial molecular generation model to generate historical drug candidates; The matching degree information between the molecular topological feature representation of the historical candidate drug and the historical unified topological feature representation is used as the reward signal; By using reinforcement learning to update the parameters of the initial molecular generation model based on the reward signal, and iteratively training until a preset convergence condition is met, a molecular generation model for generating candidate drugs is obtained.

8. A drug design device based on topological data analysis, characterized in that, include: The conformation acquisition module is used to acquire multiple three-dimensional structural conformations of the target protein and determine the binding pocket region of each three-dimensional structural conformation to construct a conformation set. The descriptor generation module is used to calculate a topological descriptor representing the topological relationship between atoms based on the atomic coordinates of the combined pocket region of each of the three-dimensional structural conformations, and to convert the topological descriptor into the corresponding conformation feature vector. The pooling module is used to perform pooling operations on the conformation feature vectors corresponding to each of the three-dimensional structural conformations to generate a unified topological feature representation that characterizes the common topological feature pattern of the target protein binding pocket under multiple three-dimensional structural conformations. A drug design module is used to screen or generate candidate drugs based on the unified topological feature representation.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the drug design method based on topology data analysis as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when executed by a processor, the computer program implements the steps of the drug design method based on topology data analysis as described in any one of claims 1 to 7.

Images