Method and related device for guiding generation of three-dimensional molecular point cloud based on similarity gradient
By extracting simplified point clouds of target molecules as conditional information, the pre-trained model is guided to generate molecules that match the target, solving the problems of high computational cost and poor flexibility in existing technologies, and achieving efficient molecule generation and improved targeting.
Patent Information
- Application Number
- CN202610412177.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-31
- Publication Date
- 2026-07-07
AI Technical Summary
Existing shape-guided molecular generation methods suffer from high computational costs, poor flexibility, and difficulty in supporting simplified or custom geometric constraints, which affect the efficiency and accuracy of molecular generation.
By extracting simplified point cloud from the three-dimensional structure of the target molecule as delineation condition information, inputting it into a pre-trained three-dimensional molecular point cloud diffusion model, and introducing point cloud similarity gradient during the inverse denoising process, the generation of molecular point cloud matching the delineation conditions is guided.
It reduces computational resource consumption and model training time, improves the efficiency and cost-effectiveness of the drug discovery process, and generates molecules with higher target binding affinity and greater diversity.
Smart Images
Figure CN122348013A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of drug molecule design technology, and relates to a method and related apparatus for generating three-dimensional molecular point clouds based on similarity gradient guidance. Background Technology
[0002] Drug development is a high-investment, high-risk, and lengthy process. To accelerate the discovery of lead compounds and reduce development costs, computational-aided drug design methods are widely used. Among these, virtual screening is a traditional mainstream strategy, using molecular docking or similarity search to screen potential candidate molecules from existing molecular libraries. However, these methods are severely limited by the size and chemical diversity of existing molecular libraries, making it difficult to generate molecules with novel structures that are highly adapted to specific target pockets.
[0003] In recent years, with the development of artificial intelligence technology, generative models, especially diffusion models, have been introduced into the field of drug design, demonstrating great potential. Unlike virtual screening, generative models can directly learn the distribution patterns of molecular structures and properties from chemical data and generate novel molecules with ideal physicochemical properties or biological activities from scratch, thereby breaking through the limitations of existing molecular libraries and exploring a broader chemical space.
[0004] In molecular generative tasks targeting specific proteins, a core challenge lies in ensuring a high degree of spatial complementarity between the three-dimensional geometry of the generated molecule and the binding pocket of the target protein. Research indicates that the effective binding of a molecule to its target depends not only on its chemical groups but also, and perhaps more importantly, on the precise matching of its overall or local spatial configuration with the topological structure of the pocket. Therefore, effectively integrating shape constraints into the generative process is crucial for enhancing the targeting and druggability of the generated molecules.
[0005] Currently, two representative works have attempted to guide molecular generation using shape information: DiffSMol proposed using 3D molecular surfaces as conditional signals. This method first trains an isotropic shape encoder to map the complete surface geometry of a reference molecule into a latent embedding, then injects it as a condition into the diffusion model, and retrains the entire generation network. Although it can generate molecules with similar surfaces, it relies on the complete molecular structure as input and requires fine-tuning of the model, limiting its flexibility and efficiency. UniGuide attempts to introduce external conditions during sampling through gradient guidance without modifying the weights of the pre-trained diffusion model. However, it supports complete reference molecules as guiding signals and cannot accept more abstract, local, or user-defined geometric constraints. The generated results are limited by the existing structure of the reference molecule and are difficult to flexibly adapt to the topological features of different target pockets.
[0006] In summary, existing shape-guided molecular generation methods suffer from two major bottlenecks: first, they rely on conditional model retraining, which is computationally expensive and lacks flexibility; second, the expression of guiding conditions is redundant and inflexible, making it difficult to support simplified or custom geometric constraints; thus affecting the efficiency and accuracy of molecular generation. Summary of the Invention
[0007] The purpose of this invention is to provide a method and related apparatus for generating three-dimensional molecular point clouds based on similarity gradient guidance, so as to solve the technical problems of low efficiency and cost of existing molecular generation methods and difficulty in supporting simplified or customized geometric constraints.
[0008] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for generating three-dimensional molecular point clouds based on similarity gradient guidance, comprising the following steps: A simplified point cloud representing the spatial contour of the target molecule is extracted from its three-dimensional structure as delineation condition information. The delineation condition information is input into a pre-trained three-dimensional molecular point cloud diffusion model; the three-dimensional molecular point cloud diffusion model includes a forward noise addition process and a reverse noise removal process; In the reverse denoising process of the three-dimensional molecular point cloud diffusion model, a similarity gradient is introduced to calculate the similarity of the point cloud to guide the sampling process to generate a molecular point cloud that matches the delineation condition information in the three-dimensional spatial configuration.
[0009] Furthermore, the step of extracting a simplified point cloud representing the spatial contour of the target molecule from its three-dimensional structure as a delineation condition information specifically includes: Identify the ring structures in the target molecule and perform average calculations on the coordinates of all atoms within each ring structure to generate a new delineation node; Delete all nodes with a degree of 1 and remove all branches; Delete the node with a degree of 2, and directly connect the two nodes connected to the deleted node; finally, the delineation condition information is obtained.
[0010] Furthermore, the expression for the forward noise addition process of the three-dimensional molecular point cloud diffusion model is as follows:
[0011] Where N is the total number of atoms in the molecule; For point cloud data, x is the coordinate of an atom, and h is the type of atom; It is a given noise scheduling table.
[0012] Furthermore, the expression for the inverse denoising process of the three-dimensional molecular point cloud diffusion model is as follows:
[0013] in, It is a conditional mean function; Standard deviation; Represents the identity matrix.
[0014] Furthermore, the step of introducing point cloud similarity calculation similarity gradient in the reverse denoising process of the three-dimensional molecular point cloud diffusion model to guide the sampling process to generate a molecular point cloud whose three-dimensional spatial configuration matches the delineation condition information specifically includes: Calculate the similarity gradient using point cloud similarity:
[0015] in, Indicates point cloud similarity; Similarity gradient; The sampling process is guided by similarity gradient:
[0016] in, To outline the conditional information; Indicates noise; The final result is a three-dimensional molecular point cloud.
[0017] Furthermore, the point cloud similarity is characterized by soft bevel distance, and the specific calculation process is as follows: Given two point clouds A and B, the chamfer distance is the sum of the distance from each point in A to its nearest neighbor in B and the distance from each point in B to its nearest neighbor in A.
[0018] Introducing a softmax weighted average to calculate the soft bevel distance:
[0019]
[0020] in, This represents the soft bevel distance between point cloud A and point cloud B; This represents the Euclidean distance between two points; This represents the temperature parameter, used to control the smoothness of the exponentially weighted average.
[0021] Furthermore, the point cloud similarity is characterized by Gaussian volume overlap, and the specific calculation process is as follows: Model each point in point cloud A and point cloud B as follows: The isotropic Gaussian function in:
[0022]
[0023]
[0024] in, Control the width of the Gaussian function; This represents the total overlap volume between point cloud A and point cloud B; This represents the overlap volume between point cloud A and itself; This represents the overlap volume between point cloud B and itself; Represents a single element in A With a single element in B Gaussian overlap contribution between them.
[0025] Secondly, the present invention provides a system for generating three-dimensional molecular point clouds based on similarity gradient guidance, comprising: The information extraction module is used to extract a simplified point cloud representing the spatial contour of the target molecule from its three-dimensional structure, as delineation condition information. The diffusion model input module is used to input the delineation condition information into the pre-trained three-dimensional molecular point cloud diffusion model; the three-dimensional molecular point cloud diffusion model includes a forward noise addition process and a reverse noise removal process; The gradient guidance module is used to introduce a similarity gradient for point cloud similarity calculation during the reverse denoising process of the three-dimensional molecular point cloud diffusion model, so as to guide the sampling process to generate a molecular point cloud that matches the delineation condition information in three-dimensional spatial configuration.
[0026] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for generating three-dimensional molecular point clouds based on similarity gradients as described above.
[0027] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for generating a three-dimensional molecular point cloud based on similarity gradient.
[0028] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a method and related apparatus for generating 3D molecular point clouds based on similarity gradient guidance. By extracting simplified delineated point clouds of the target molecule as delineation condition information, this delineated point cloud information is introduced into the inverse sampling process of a pre-trained unconditional 3D molecular point cloud diffusion model, without requiring modification of model weights or retraining. This overcomes the limitation of existing technologies that require adding additional conditional information and retraining the model on top of the original generation model. It significantly reduces computational resource consumption and model training time, improving the efficiency and cost-effectiveness of drug discovery. It also allows for rapid transfer to more advanced unconditional molecular generation models once they become available. Attached Figure Description
[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0030] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system of the present invention; Figure 3 This is the extraction process of molecular delineation in an embodiment of the present invention; Figure 4 These are two point cloud similarity calculation metrics in embodiments of the present invention; Figure 5 This is a schematic diagram illustrating the outlining and guidance during the sampling of the diffusion model in an embodiment of the present invention; Figure 6 This is an example of molecular generation with shape constraints according to an embodiment of the present invention; Figure 7 The interaction diagram between the generated molecule and the target protein is shown in the embodiment of the present invention. Detailed Implementation
[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present 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 application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.
[0032] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that the embodiments of this application 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 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.
[0033] It should be understood that in the embodiments of this application, "at least one" means one or more, and "more than one" means two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship. "Contains A, B and / or C" means containing any one, two, or three of A, B, and C.
[0034] It should be understood that in the embodiments of this application, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.
[0035] This invention provides a method for generating 3D molecular point clouds based on similarity gradient guidance, which can be executed by an electronic device, such as a terminal or a server. The terminal can be a smartphone, tablet, laptop, or other similar device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. It is understood that this invention does not limit the specific entity executing this method for generating 3D molecular point clouds based on similarity gradient guidance.
[0036] The technical solution of this application will be described in detail below through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments described below are used to explain the technical solution of this application and are not intended to limit actual use.
[0037] See Figure 1 This invention provides a method for generating three-dimensional molecular point clouds based on similarity gradient guidance. By extracting concise and crucial delineation information from the three-dimensional structure of the target molecule as a condition, and using differentiable point cloud similarity metrics (such as soft chamfer distance and Gaussian volume overlap) to calculate the conditional gradient, precise guidance of the diffusion process is achieved, thereby generating a delineated structure with a similar contour to the target molecule. Specifically, the method includes the following steps: S1, extracts a simplified point cloud representing the spatial contour of the target molecule from its three-dimensional structure, as delineation condition information; To enhance the diversity of the generated molecules, this invention extracts key features from the sample molecule as outlines before guiding the process. This ensures that the generated molecules are similar to the sample molecule only in their contours, avoiding excessive 2D structural resemblance that would result in too few new molecules.
[0038] like Figure 3 First, each ring structure is identified and processed. The coordinates of all atoms in the ring are averaged to obtain a new delineation node (ring point) to represent the ring structure.
[0039] Then delete all nodes with a degree of 1 (connected to only one node) and all branches. The nodes generated in the first step are retained because their information is crucial to the molecular information.
[0040] Finally, delete the nodes with a degree of 2 (connecting only two nodes), and connect the two nodes connected to them to make the entire outline more concise and smooth. Similarly, the nodes in this step are also retained.
[0041] Finally, simplified molecular delineation point cloud information that can be used to guide diffusion models is obtained.
[0042] S2, input the delineation condition information into the pre-trained three-dimensional molecular point cloud diffusion model; the three-dimensional molecular point cloud diffusion model includes a forward noise addition process and a reverse noise removal process; In the unconditional 3D molecular point cloud diffusion generation model, molecular information is expressed in the form of point clouds: Where x is the coordinate of an atom, h is the type of atom (existing as a one-hot vector), and N is the total number of atoms in the molecule. Point cloud data. The diffusion process involves concatenating the atomic coordinates with the one-hot vectors of the atomic types.
[0043] The diffusion model learns a Markov chain, which involves a forward process that adds noise to data from a distribution q(z), and then learns to reverse this process to generate new samples from a normal distribution. This is given by data points sampled from the true underlying distribution. Forward process Gaussian noise will be added gradually:
[0044] in, It is a given noise schedule that controls the degree of noise addition at each step, ensuring that the noise is eventually converted to a normal distribution.
[0045] In the actual reverse engineering process Raw data is needed. The diffusion model neural network predicts noise, yielding the following results:
[0046] Sampling starts from Gaussian noise Start, use gradually Sampling was performed to obtain entirely new point cloud data. This point cloud data predicts bonding relationships based on the distance between points and the type of atoms, thus deriving a complete new molecule.
[0047] The above outlines the basic theory of training and sampling for an unconditional 3D molecular point cloud diffusion generation model. Without altering the existing trained network, such as... Figure 5 Conditional information can be introduced into a neural network using the following formula:
[0048] In traditional classifier-guided diffusion models, It is a classifier because condition c is a discrete label, while in this invention, condition c is information that delineates the point cloud, which is continuous information. The similarity between point clouds is used to fit the data. :
[0049] S3, In the reverse denoising process of the three-dimensional molecular point cloud diffusion model, a similarity gradient is introduced to calculate the similarity of the point cloud to guide the sampling process to generate a molecular point cloud that matches the delineation condition information in the three-dimensional spatial configuration.
[0050] See Figure 4 This invention employs two point cloud similarity calculation methods. , used to calculate the guiding gradient.
[0051] A. Soft chamfer distance Given two point clouds A and B, the chamfer distance is the sum of the distance from each point in A to its nearest neighbor in B and the distance from each point in B to its nearest neighbor in A.
[0052] Traditional chamfer distance (CD) is non-differentiable due to the inclusion of a min operation. This invention introduces a softmax weighted average to replace the min operation, achieving a differentiable soft chamfer distance:
[0053]
[0054] B. Gaussian volume overlap The Gaussian description of point clouds is used to calculate the volume overlap as the similarity between point clouds, and each point in point cloud A and point cloud B is modeled as follows: isotropic Gaussian function in
[0055]
[0056]
[0057] in, Control the width of the Gaussian function.
[0058] These two similarity measures are used to represent the above. This allows for sampling generation guided by diffusion based on the similarity gradient between point clouds.
[0059] See Figure 2This invention discloses a system for generating 3D molecular point clouds based on similarity gradient guidance, comprising an information extraction module, a diffusion model input module, and a gradient guidance module. The information extraction module extracts simplified point clouds representing the spatial contours of the target molecule from its 3D structure, serving as delineation condition information. The diffusion model input module inputs the delineation condition information into a pre-trained 3D molecular point cloud diffusion model, which includes a forward noise addition process and a reverse noise reduction process. The gradient guidance module introduces a similarity gradient calculation during the reverse noise reduction process of the 3D molecular point cloud diffusion model to guide the sampling process in generating molecular point clouds whose 3D spatial configuration matches the delineation condition information. This invention utilizes a gradient guidance mechanism to introduce delineation point cloud information into the reverse sampling process of a pre-trained unconditional 3D molecular point cloud diffusion model. This overcomes the limitation of existing technologies that require adding additional condition information and retraining the model on top of the original generation model. It significantly reduces computational resource consumption and model training time, improving the efficiency and cost-effectiveness of the drug discovery process. It can also be used for rapid migration once a better unconditional molecular generation model emerges.
[0060] Example: 1. SKG-guided generation improves shape similarity while maintaining diversity. This embodiment uses 1000 sample molecules from the MOSES test set, with each sample molecule generating 100 new molecules. The unconditional molecule generation model (EDM) used in this invention is also retrained using the MOSES training set.
[0061] Table 1: Comparison of Shape Similarity and Diversity
[0062] As shown in Table 1, compared to DiffSMol, based on SKG Chamfer Distance-guided graphs achieved the highest performance in average shape similarity (0.763 vs. 0.738) while maintaining similar graph structure similarity (average simg ≈ 0.23).
[0063] For unconditional EDM, skeleton guidance brings a more significant improvement: EDM + SKG Chamfer The average shape similarity was increased from 0.539 to 0.734, achieving performance comparable to the retrained conditional model DiffSMol (0.738) without additional training, relying solely on sampling-stage guidance. This demonstrates that effective shape control does not depend on explicit conditional retraining.
[0064] In the comparison of different guidance objectives, SKG ChamferGuiding focuses more on shape fidelity; while Overlap guidance sacrifices slightly reduced shape similarity for higher diversity and lower graph structure similarity. EDM+Overlap achieved the highest diversity (0.794) and the lowest average simg (0.185). This phenomenon reflects the trade-off between point-level and volumetric geometric constraints in terms of fidelity and diversity.
[0065] like Figure 6 As shown, the SKG used in this invention Chamfer The molecules generated by Uniguide have a higher shape similarity than those generated by Uniguide because Uniguide can only constrain the generated molecules to a shape and cannot directly use the original information of the sample molecules.
[0066] 2. In tasks based on protein pocket generation, molecules generated under SKG guidance have stronger virtual screening capabilities.
[0067] Given that recent studies have shown that the correlation between ligand-based virtual screening metrics and actual experimental hit rates is significantly higher than that of docking scores, we adopted BEDROC and EF@1 as the primary evaluation metrics. This embodiment uses the DUDE dataset to guide the generation of a conditional diffusion model based on protein pocket generation, DiffSBDD. A total of 100 protein pocket-ligand pairs were used, with 100 molecules sampled from each pair.
[0068] Table 2: Evaluation Results of Virtual Screening Capability
[0069] As shown in Table 2, the performance improvement brought by the guidance strategy is particularly significant when evaluated using structure-aware or pocket-aligned encoders such as DrugCLIP and UniMol; in contrast, the screening results based on 2D fingerprints show only limited improvement. This difference indicates that the guidance mechanism mainly optimizes the 3D conformation of the generated molecule and its interaction-related features, without significantly altering the underlying molecular graph topology. Therefore, similarity measures relying solely on 2D connectivity information are not sensitive to the improvements brought by the guidance strategy. Figure 7 As shown, the SKG used in this invention Overlap When applied to the protein pocket-based generation model DiffSBDD, the non-covalent bond forces between the guided molecules and the protein pockets are more consistent with the original molecules compared to the molecules generated without guidance in the original model.
[0070] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a method for generating three-dimensional molecular point clouds based on similarity gradient guidance.
[0071] This invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for generating three-dimensional molecular point clouds based on similarity gradients in the above embodiments.
[0072] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0073] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for generating three-dimensional molecular point clouds based on similarity gradient guidance, characterized in that, Includes the following steps: A simplified point cloud representing the spatial contour of the target molecule is extracted from its three-dimensional structure as delineation condition information. The delineation condition information is input into a pre-trained three-dimensional molecular point cloud diffusion model; the three-dimensional molecular point cloud diffusion model includes a forward noise addition process and a reverse noise removal process; In the reverse denoising process of the three-dimensional molecular point cloud diffusion model, a similarity gradient is introduced to calculate the similarity of the point cloud to guide the sampling process in generating a molecular point cloud that matches the delineation condition information in the three-dimensional spatial configuration.
2. The method for generating three-dimensional molecular point clouds based on similarity gradient guidance according to claim 1, characterized in that, The step of extracting a simplified point cloud representing the spatial contour of the target molecule from its three-dimensional structure as a delineation condition information specifically includes: Identify the ring structures in the target molecule and perform average calculations on the coordinates of all atoms within each ring structure to generate a new delineation node; Delete all nodes with a degree of 1 and remove all branches; Delete the node with a degree of 2, and directly connect the two nodes connected to the deleted node; finally, the delineation condition information is obtained.
3. The method for generating three-dimensional molecular point clouds based on similarity gradient guidance according to claim 1, characterized in that, The expression for the forward noise addition process of the three-dimensional molecular point cloud diffusion model is as follows: Where N is the total number of atoms in the molecule; For point cloud data, x is the coordinate of an atom, and h is the type of atom; It is a given noise scheduling table.
4. The method for generating three-dimensional molecular point clouds based on similarity gradient guidance according to claim 1, characterized in that, The expression for the inverse denoising process of the three-dimensional molecular point cloud diffusion model is as follows: in, It is a conditional mean function; Standard deviation; Represents the identity matrix.
5. The method for generating three-dimensional molecular point clouds based on similarity gradient guidance according to claim 1, characterized in that, The step of introducing point cloud similarity calculation similarity gradient in the reverse denoising process of the three-dimensional molecular point cloud diffusion model to guide the sampling process to generate a molecular point cloud whose three-dimensional spatial configuration matches the delineation condition information specifically includes: Calculate the similarity gradient using point cloud similarity: in, Indicates point cloud similarity; Similarity gradient; The sampling process is guided by similarity gradient: in, To outline the conditional information; Indicates noise; The final result is a three-dimensional molecular point cloud.
6. The method for generating three-dimensional molecular point clouds based on similarity gradient guidance according to claim 1, characterized in that, The point cloud similarity is represented by soft bevel distance, and the specific calculation process is as follows: Given two point clouds A and B, the chamfer distance is the sum of the distance from each point in A to its nearest neighbor in B and the distance from each point in B to its nearest neighbor in A. Introducing a softmax weighted average to calculate the soft bevel distance: in, This represents the soft bevel distance between point cloud A and point cloud B; This represents the Euclidean distance between two points; This represents the temperature parameter, used to control the smoothness of the exponentially weighted average.
7. The method for generating three-dimensional molecular point clouds based on similarity gradient guidance according to claim 1, characterized in that, The point cloud similarity is characterized by Gaussian volume overlap, and the specific calculation process is as follows: Model each point in point cloud A and point cloud B as follows: The isotropic Gaussian function in: in, Control the width of the Gaussian function; This represents the total overlap volume between point cloud A and point cloud B; This represents the overlap volume between point cloud A and itself; This represents the overlap volume between point cloud B and itself; Represents a single element in A With a single element in B Gaussian overlap contribution between them.
8. A system for generating three-dimensional molecular point clouds based on similarity gradient guidance, characterized in that, include: The information extraction module is used to extract a simplified point cloud representing the spatial contour of the target molecule from its three-dimensional structure, as delineation condition information. The diffusion model input module is used to input the delineation condition information into the pre-trained three-dimensional molecular point cloud diffusion model; the three-dimensional molecular point cloud diffusion model includes a forward noise addition process and a reverse noise removal process; The gradient guidance module is used to introduce a similarity gradient for point cloud similarity calculation during the reverse denoising process of the three-dimensional molecular point cloud diffusion model, so as to guide the sampling process to generate a molecular point cloud that matches the delineation condition information in three-dimensional spatial configuration.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for generating three-dimensional molecular point clouds based on similarity gradient as described in any one of claims 1-7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for generating three-dimensional molecular point clouds based on similarity gradient as described in any one of claims 1-7.