Physical and chemical knowledge guided multi-resolution molecular property prediction method and system
By introducing molecular potential energy and functional group information, molecular structure is fully encoded, solving the problem of lack of physical and chemical information in existing technologies, and achieving high-quality molecular property prediction and enhanced interpretability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG NORMAL UNIV
- Filing Date
- 2024-09-27
- Publication Date
- 2026-07-31
AI Technical Summary
Existing molecular property prediction techniques lack comprehensive consideration of physical and chemical information, resulting in insufficient physicochemical interpretability of the models. They also ignore various intramolecular interaction forces and local structural information, leading to poor prediction performance.
A multi-granularity molecular property prediction method guided by physical and chemical knowledge is adopted. It comprehensively encodes molecular structure through edge granularity, point granularity, subgraph granularity and graph granularity information, and uses molecular potential energy and functional group information for embedding to generate high-quality and interpretable molecular characterization.
It achieves comprehensive and high-quality expression of molecular property predictions, enhances the physical and chemical interpretability of the model, and improves prediction performance.
Smart Images
Figure CN119152968B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of molecular representation learning and molecular property prediction technology, and particularly relates to a multi-granularity molecular property prediction method and system guided by knowledge from the fields of physics and chemistry. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] The purpose of molecular property prediction is to accurately predict the properties of small molecules in the screening library, so as to facilitate the selection of potential drug candidates and reduce the time and cost of drug development.
[0004] The inventors discovered that current molecular property prediction techniques suffer from a lack of consideration for this type of data. Molecular data itself contains a wealth of physical and chemical information. From a molecular dynamics perspective, physical information is crucial for ensuring molecular stability. Furthermore, certain chemical information significantly influences molecular properties. However, current molecular property prediction techniques fail to take this information into account, resulting in insufficient physicochemical interpretability of the models. In addition, existing techniques focus only on encoding atoms within the molecule, neglecting other granularities and the various interactions within the molecule. Moreover, the limited message passing mechanisms of existing techniques cannot adequately embed local structural information of the molecule, and global structural information is often lost; all of these problems contribute to poor molecular prediction performance. Summary of the Invention
[0005] To address at least one of the technical problems existing in the aforementioned background art, this invention provides a multi-granularity molecular property prediction method and system guided by knowledge from the fields of physics and chemistry. This method uses physical and chemical knowledge as guidance and simultaneously considers edge-granularity (chemical bonds), point-granularity (atoms), subgraph-granularity (functional groups), and graph-granularity (molecular graph) information to comprehensively encode various hierarchical structures of molecules. Specifically, molecular potential energy is used as physical information to propagate at the edge and point-granular levels, while functional groups are used as chemical information to achieve full embedding at the subgraph-granularity level. Finally, the outputs of the above granularities are aggregated at the graph-granular level to generate high-quality and interpretable molecular characterizations, which are then applied to downstream molecular property prediction tasks.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] The first aspect of this invention provides a method for predicting multi-granularity molecular properties guided by knowledge from the fields of physics and chemistry, comprising the following steps:
[0008] Obtain multi-granularity information of the molecule to be predicted;
[0009] Combining the physical and chemical knowledge of the molecules to be predicted, and based on the multi-granularity information of the molecules to be predicted and the constructed molecular property prediction model, molecular characterization is obtained for molecular property prediction.
[0010] The process of constructing the molecular property prediction model includes:
[0011] Treating chemical bonds within a molecule as edges in a molecular graph and atoms within a molecule as points in the molecular graph, we introduce molecular potential energy. Based on the energy of the interaction forces corresponding to each potential term in the molecular potential energy, we calculate the relative importance of each hop neighbor of each edge or point in the molecular graph and update the embedding representation of edges and points in the molecular graph.
[0012] Each functional group is considered as a subgraph in the molecular graph and compressed into subgraph nodes. The 3D distance between subgraph nodes is used as the bias, and the interrelationships between subgraph nodes are considered to update the representation of the subgraph nodes.
[0013] Molecular characterization is obtained based on the embedded representation of edges and points in the molecular graph, the representation of subgraph nodes, and the original molecular features.
[0014] Furthermore, the introduction of molecular potential energy, based on the energy of the interaction forces corresponding to each potential energy term included in the molecular potential energy, calculates the relative importance of each hop neighbor of each edge or vertex in the molecular graph, and updates the embedding representation of edges and vertices in the molecular graph, specifically including:
[0015] Calculate the bond lengths, bond angles, dihedral angles, and non-bond distances of a molecule based on the coordinates of atoms within the molecule;
[0016] The bond lengths, bond angles, dihedral angles, and non-bond distances of molecules are embedded in an OPLS force field to generate corresponding stretching energy, bending energy, torsional energy, and non-bond interaction energy.
[0017] The relative importance of 1-2 jump adjacent edges of each edge is calculated based on bending energy and torsional energy;
[0018] The relative importance of 1-3 hop neighboring points of each point is calculated based on stretching energy, bending energy, torsional energy and nonbonded interaction energy.
[0019] The embedding representations of edges and points in the molecular graph are updated based on the relative importance of the 1-2 hop neighbors of each edge and the relative importance of the 1-3 hop neighbors of each point.
[0020] Furthermore, the calculation of the relative importance of 1-2 hop neighboring edges of each edge based on bending energy and torsional energy includes:
[0021] Each edge updates its representation by aggregating information from direct connection points and 1-2 hop adjacent edges. The embedding of direct connection points is directly passed to the target edge. The bond angle curvature between the 1-hop adjacent edge and the target edge can be used as the message passing weight of the 1-hop adjacent edge. The dihedral angle twist between the 2-hop adjacent edge and the target edge can be used as the message passing weight of the 2-hop adjacent edge.
[0022] Furthermore, the weight of a 1-hop neighbor is inversely proportional to the bond angle, and the weight of a 2-hop neighbor is inversely proportional to the dihedral angle. The larger the bond angle or dihedral angle, the higher the bending energy or torsional energy, and the lower the weight assigned to the corresponding neighbor by the target edge.
[0023] Furthermore, in the process of assigning neighbor weights, if both are 1-hop neighbors, the neighbor with the larger bond angle is assigned a lower weight to the target edge; if both are 2-hop neighbors, the neighbor with the larger dihedral angle is assigned a lower weight to the target edge.
[0024] Furthermore, the calculation of the relative importance of 1-3 hop neighbors of each point based on stretching energy, bending energy, torsional energy, and nonbonded interaction energy includes:
[0025] Each point updates its own representation by aggregating information from direct connecting edges and 1-3 hop neighboring points. The embedding representation of the direct connecting edges is directly passed to the target point. The bond length stretching energy between the 1-hop neighboring point and the target point is used as the aggregation weight of the 1-hop neighboring point. The bond angle bending energy and non-bond interaction energy between the 2-hop neighboring point and the target point are used as the aggregation weight of the 2-hop neighboring point. The dihedral angle torsion energy and non-bond interaction energy between the 3-hop neighboring point and the target point are used as the aggregation weight of the 3-hop neighboring point.
[0026] Furthermore, the weight of a 1-hop neighbor to the target point is inversely proportional to the bond length, the weight of a 2-hop neighbor to the target point is inversely proportional to the bond angle and the distance between points, and the weight of a 3-hop neighbor to the target point is inversely proportional to the dihedral angle and the distance between points.
[0027] A second aspect of the present invention provides a multi-granularity molecular property prediction system guided by knowledge in the fields of physics and chemistry, comprising:
[0028] The information acquisition module is used to acquire multi-granularity information of the molecule to be predicted.
[0029] The molecular property prediction module combines the physical and chemical knowledge of the molecule to be predicted, and obtains molecular characterization for molecular property prediction based on the multi-granularity information of the molecule to be predicted and the constructed molecular property prediction model.
[0030] The process of constructing the molecular property prediction model includes:
[0031] Treating chemical bonds within a molecule as edges in a molecular graph and atoms within a molecule as points in the molecular graph, we introduce molecular potential energy. Based on the energy of the interaction forces corresponding to each potential term in the molecular potential energy, we calculate the relative importance of each hop neighbor of each edge or point in the molecular graph and update the embedding representation of edges and points in the molecular graph.
[0032] Each functional group is considered as a subgraph in the molecular graph and compressed into subgraph nodes. The 3D distance between subgraph nodes is used as the bias, and the interrelationships between subgraph nodes are considered to update the representation of the subgraph nodes.
[0033] Molecular characterization is obtained based on the embedded representation of edges and points in the molecular graph, the representation of subgraph nodes, and the original molecular features.
[0034] A third aspect of the present invention provides a computer-readable storage medium.
[0035] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the multi-granularity molecular property prediction method guided by knowledge in the fields of physics and chemistry as described above.
[0036] A fourth aspect of the present invention provides a computer device.
[0037] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps in the knowledge-guided multi-granularity molecular property prediction method described above in the fields of physics and chemistry.
[0038] Compared with the prior art, the beneficial effects of the present invention are:
[0039] This invention provides a multi-granularity molecular property prediction method and system guided by knowledge from the fields of physics and chemistry. It comprehensively integrates information from edges, nodes, subgraphs, and the graph within a molecular graph, encodes them through different granularity modules, and enables interaction, thereby achieving a comprehensive and high-quality molecular expression capability of the model. Based on the important physical concept of "molecular potential energy" and the important chemical concept of "functional group," they are introduced into message passing networks and self-attention mechanisms, respectively. This achieves superior performance while also complementing the local and global structural information within the molecule, increasing the physical and chemical interpretability of the model's prediction results.
[0040] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0041] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0042] Figure 1 This is a flowchart of a multi-granularity molecular property prediction method guided by knowledge in the fields of physics and chemistry, according to an embodiment of the present invention.
[0043] Figure 2 This is a flowchart of molecular expression guided by physical domain knowledge in an embodiment of the present invention;
[0044] Figure 3 This is a flowchart of molecular expression guided by chemical knowledge in an embodiment of the present invention;
[0045] Figure 4 This is a schematic diagram of the structure of a multi-granularity molecular property prediction system guided by knowledge in the fields of physics and chemistry, according to an embodiment of the present invention. Detailed Implementation
[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0047] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0048] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0049] This invention provides a multi-granularity molecular property prediction method and system guided by knowledge from the fields of physics and chemistry. The physical information is molecular potential energy, and the chemical information is functional groups. The multi-granularity includes: chemical bonds, atoms, functional groups, and a molecular graph. First, it is necessary to obtain the characteristics of all chemical bonds and atoms of the molecule whose properties are to be predicted, the coordinates of all atoms, the characteristics and coordinates of all functional groups, and the molecular characteristics. Second, the characteristics of chemical bonds and atoms, as well as the coordinates of atoms, are input into the edge-granularity module and the point-granularity module, with the molecular potential energy serving as guiding information. The characteristics and coordinates of functional groups are input into the sub-graph granularity module. Finally, the molecular characteristics and the outputs of the above modules are summarized in the graph granularity module to generate a molecular characterization used for property prediction.
[0050] Example 1
[0051] like Figure 1-2 As shown, this embodiment provides a multi-granularity molecular property prediction method guided by knowledge from the fields of physics and chemistry, including the following steps:
[0052] S101: Obtain multi-granularity information of the molecule to be predicted;
[0053] The multi-granularity information of the molecule to be predicted includes the features of all chemical bonds and atoms, as well as the coordinates of all atoms, the features and coordinates of all functional groups, and the features of the molecule.
[0054] The process of obtaining the features and coordinates of all functional groups is as follows:
[0055] S1011: Predefine SMART identifiers for all functional groups, and use them to retrieve all functional groups in the molecule by subgraph matching;
[0056] S1012: Initialize the functional groups based on their chemical characteristics;
[0057] S1013: The coordinates of a functional group are calculated based on the average of the coordinates of all atoms contained in that functional group.
[0058] S102: Guided by physical and chemical knowledge, this module treats chemical bonds within a molecule as edges of a molecular graph, and atoms as points within the graph. Chemical bonds are input into the edge-grained module, while atomic features and coordinates are input into the point-grained module. Next, the features and coordinates of functional groups are input into the sub-graph-grained module. Finally, the outputs of all the above modules and the molecular features are summarized in the graph-grained module to form a molecular characterization for property prediction.
[0059] like Figure 3 As shown, the potential energy term in molecular potential energy is used to assist in the calculation of aggregate weights during edge and vertex message passing. Specific details include the following:
[0060] First, bond lengths, bond angles, dihedral angles, and non-bond distances are calculated based on the coordinates of the atoms.
[0061] Secondly, this geometric information is processed through the OPLS force field to generate corresponding stretching energy, bending energy, torsional energy, and non-bonded interaction energy. The sum of these four energies constitutes the expression of molecular potential energy under the OPLS field.
[0062] Finally, this energy information is used to assist in weight calculation during message passing between the edge-granularity and point-granularity modules. The expression for molecular potential energy and its embedded expressions for the four types of energy it contains are as follows:
[0063] E = E bond +E angle +Etorsion +E nonbond ,
[0064] E bond =[r,r 2 ],
[0065] E angle =[θ,θ 2 ],
[0066] E torsion =[cosφ, cos2φ, cos3φ, sinφ, sin2φ, sin3φ],
[0067] E nonbond =[r -1 ,r -12 ,r -6 ],
[0068] Among them, E bond E represents the stretching energy in molecular potential energy. angle E represents the bending energy in molecular potential energy. torsion E represents the torsional energy in molecular potential energy. nonbond The nonbonded interaction energy in molecular potential energy is represented by r, which represents the bond length (interatomic distance), θ, which represents the bond angle, φ, which represents the dihedral angle, and [·] which represents the bonding operation.
[0069] The edge granularity module takes the features of edges and vertices as input to update the embedded representation of edges. Each edge in the molecular graph aggregates information from its neighboring vertices or edges.
[0070] Specifically, each edge updates its representation by aggregating information from its direct connection points and 1-2 hop neighboring edges. The embedding of the direct connection point represents the direct transmission to the target edge, the bond angle curvature between the 1-hop neighboring edge and the target edge serves as the message passing weight for the 1-hop neighboring edge, and the dihedral torsion between the 2-hop neighboring edge and the target edge serves as the message passing weight for the 2-hop neighboring edge.
[0071] The specific formula is as follows:
[0072]
[0073]
[0074] Where [·] represents a join operation, h uv The characteristic of an edge, h u The characteristics of the points are represented by u, v, and h. v The characteristics of the points are represented by l, where l represents the depth of the network, and m represents the network depth. one_hop_edge This represents the message summary where the target edge originates from a 1-hop adjacent edge, m. two_hop_edgeh represents the message summary from the target edge to the two adjacent edges. g Indicates the characteristics of a molecule, E angle It is a mapping function of bond angle to bending energy, E dihedral It is a mapping function from dihedral angle to torsional energy. Angle() is a function to calculate the bond angle, and Dihedral() is a function to calculate the dihedral angle.
[0075] The point-granularity module takes the features of points and edges as input to update the embedding representation of points. Each point within the molecular graph aggregates information from its neighboring edges or points. Specifically, each point updates its representation by aggregating information from directly connected edges and 1-3 hop neighbor points. The embedding representation of directly connected edges is directly passed to the target point. The bond length stretching energy between the 1-hop neighbor and the target point serves as the aggregation weight for the 1-hop neighbor point; the bond angle bending energy and non-bonded interaction energy between the 2-hop neighbor and the target point serve as the aggregation weight for the 2-hop neighbor point; and the dihedral angle torsion energy and non-bonded interaction energy between the 3-hop neighbor and the target point serve as the aggregation weight for the 3-hop neighbor point.
[0076] The specific formula is expressed as follows:
[0077]
[0078] Where, m one_hop_node This represents the message summary from the target point to its 1-hop neighbor, m. two_hop_node This represents the message summary from the target point to its two hop neighbors, m. three_hop_node This indicates that the target point originates from a message aggregated from 3 adjacent points. `Bond()` is the function to calculate bond lengths, and `Nonbond()` is the function to calculate non-bond distances (interatomic distances). E bond It is a mapping function of bond length to stretch energy, E nonbond It is a mapping function from nonbond distance to nonbond interaction energy.
[0079] In the message passing process of the above point-granularity module, the embedded representation of the directly connected edge is passed to the target point as a direct message, and the embedded representation of the 1-3 hop adjacent points is passed to the target point as an indirect message. The molecular features also provide global guidance information.
[0080] like Figure 4 As shown, retrieving and encoding functional groups within molecules improves model performance while increasing the chemical interpretability of molecular characterization and prediction results. Specific details include the following:
[0081] First, a SMART identifier is predefined for each functional group, and a subgraph matching method is used to retrieve the functional groups present in the molecule. Then, each functional group is treated as a subgraph in the molecular graph and compressed into subgraph nodes.
[0082] The subgraph granularity module takes the features and coordinates of subgraph nodes as input to update the embedding representation of the subgraph. Each subgraph node's embedding representation is updated based on its relationship with all other subgraph nodes. Specifically, a novel multi-head self-attention mechanism is defined, using the 3D distance between subgraph nodes as a bias. Based on this principle, the interrelationships between subgraph nodes are considered to update the representation of each subgraph node.
[0083] The specific formula is as follows:
[0084]
[0085] MultiHead(Q,K,V)=Concat(head1,...,head h W O ,
[0086] head k =Attention(x) s Q k_i ,x s K k_j ,x s V k_j ),
[0087] Among them, Q k_i For querying, K k_j For index, V k_j For content, x s Characteristic of functional groups, W Q W K W V and W O Let q represent the learnable matrix. ij The 3D distance between functional groups i and j is represented by d, where d represents the feature dimension of the functional group, and head is the 3D distance between functional groups i and j. h This indicates the h-th attention head.
[0088] By visualizing the weights in the message passing process of the edge-granularity module, the relationship between weights and bond angles and dihedral angles was analyzed. Through reasoning about the relationship between weights and angles in the message passing process of the point-granularity module, the relationship between message aggregation weights and energy in our model's message passing process was further determined.
[0089] Specifically, for the edge granularity module, the relationship between the aggregated weights of all 1-hop neighbors of each edge and the bond angle, and the relationship between the aggregated weights of all 2-hop neighbors of each edge and the dihedral angle, are visualized. The study found that the weight of a 1-hop neighbor is inversely proportional to the bond angle, and the weight of a 2-hop neighbor is inversely proportional to the dihedral angle. This indicates that the larger the bond angle or dihedral angle, the higher the bending or torsional energy, and the lower the weight assigned to the corresponding neighbor by the target edge. This aligns with the principles of physics; the larger the angle, the higher the energy, and the less stable the molecule. In the process of assigning neighbor weights in the model, if both are 1-hop neighbors, the neighbor with the larger bond angle is assigned a lower weight by the target edge because a larger angle means higher energy, making it less stable than other 1-hop neighbors and thus receiving a lower weight. If both are 2-hop neighbors, the neighbor with the larger dihedral angle is assigned a lower weight by the target edge, for the same reason as with 1-hop neighbors.
[0090] Furthermore, based on the above conclusions, the weight allocation of each hop neighbor in the point granularity module message transmission process is inferred: the weight of a 1-hop neighbor to the target point is inversely proportional to the bond length; the weight of a 2-hop neighbor to the target point is inversely proportional to the bond angle and the distance between points; and the weight of a 3-hop neighbor to the target point is inversely proportional to the dihedral angle and the distance between points.
[0091] Example 2
[0092] This embodiment provides a multi-granularity molecular property prediction system guided by knowledge from the fields of physics and chemistry, including:
[0093] The information acquisition module is used to acquire multi-granularity information of the molecule to be predicted.
[0094] The molecular property prediction module combines the physical and chemical knowledge of the molecule to be predicted, and obtains molecular characterization for molecular property prediction based on the multi-granularity information of the molecule to be predicted and the constructed molecular property prediction model.
[0095] The process of constructing the molecular property prediction model includes:
[0096] Treating chemical bonds within a molecule as edges in a molecular graph and atoms within a molecule as points in the molecular graph, we introduce molecular potential energy. Based on the energy of the interaction forces corresponding to each potential term in the molecular potential energy, we calculate the relative importance of each hop neighbor of each edge or point in the molecular graph and update the embedding representation of edges and points in the molecular graph.
[0097] Each functional group is considered as a subgraph in the molecular graph and compressed into subgraph nodes. The 3D distance between subgraph nodes is used as the bias, and the interrelationships between subgraph nodes are considered to update the representation of the subgraph nodes.
[0098] Molecular characterization is obtained based on the embedded representation of edges and points in the molecular graph, the representation of subgraph nodes, and the original molecular features.
[0099] Example 3
[0100] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the multi-granularity molecular property prediction method guided by knowledge in the fields of physics and chemistry as described above.
[0101] Example 4
[0102] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the multi-granularity molecular property prediction method guided by knowledge in the fields of physics and chemistry as described above.
[0103] 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 hardware embodiments, software embodiments, or embodiments 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 and optical storage) containing computer-usable program code.
[0104] 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.
[0105] 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.
[0106] 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.
[0107] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A multi-granularity molecular property prediction method guided by knowledge from the fields of physics and chemistry, characterized in that, Includes the following steps: Obtain multi-granularity information of the molecule to be predicted; Combining the physical and chemical knowledge of the molecules to be predicted, and based on the multi-granularity information of the molecules to be predicted and the constructed molecular property prediction model, molecular characterization is obtained for molecular property prediction. The process of constructing the molecular property prediction model includes: Treating chemical bonds within a molecule as edges in a molecular graph and atoms within a molecule as points in the molecular graph, we introduce molecular potential energy. Based on the energy of the interaction forces corresponding to each potential term in the molecular potential energy, we calculate the relative importance of each hop neighbor of each edge or point in the molecular graph and update the embedding representation of edges and points in the molecular graph. Each functional group is considered as a subgraph in the molecular graph and compressed into subgraph nodes. The 3D distance between subgraph nodes is used as the bias, and the interrelationships between subgraph nodes are considered to update the representation of the subgraph nodes. Molecular characterization is obtained based on the embedded representation of edges and points in the molecular graph, the representation of subgraph nodes, and the original molecular features.
2. The method for predicting multi-granularity molecular properties guided by knowledge in the fields of physics and chemistry as described in claim 1, characterized in that, The introduction of molecular potential energy, based on the energy of the interaction forces corresponding to each potential energy term contained in the molecular potential energy, calculates the relative importance of each hop neighbor of each edge or vertex in the molecular graph, and updates the embedding representation of edges and vertices in the molecular graph, specifically including: Calculate the bond lengths, bond angles, dihedral angles, and non-bond distances of a molecule based on the coordinates of atoms within the molecule; The bond lengths, bond angles, dihedral angles, and non-bond distances of molecules are embedded in an OPLS force field to generate corresponding stretching energy, bending energy, torsional energy, and non-bond interaction energy. The relative importance of 1-2 jump adjacent edges of each edge is calculated based on bending energy and torsional energy; The relative importance of 1-3 hop neighboring points of each point is calculated based on stretching energy, bending energy, torsional energy and nonbonded interaction energy. The embedding representations of edges and points in the molecular graph are updated based on the relative importance of the 1-2 hop neighbors of each edge and the relative importance of the 1-3 hop neighbors of each point.
3. The method for predicting multi-granularity molecular properties guided by knowledge in the fields of physics and chemistry as described in claim 2, characterized in that, The calculation of the relative importance of 1-2 jump adjacent edges of each edge based on bending energy and torsional energy includes: Each edge updates its representation by aggregating information from direct connection points and 1-2 hop adjacent edges. The embedding of direct connection points is directly passed to the target edge. The bond angle curvature between the 1-hop adjacent edge and the target edge can be used as the message passing weight of the 1-hop adjacent edge. The dihedral angle twist between the 2-hop adjacent edge and the target edge can be used as the message passing weight of the 2-hop adjacent edge.
4. The method for predicting multi-granularity molecular properties guided by knowledge in the fields of physics and chemistry as described in claim 3, characterized in that, The weight of a 1-jump neighbor is inversely proportional to the bond angle, and the weight of a 2-jump neighbor is inversely proportional to the dihedral angle. The larger the bond angle or dihedral angle, the higher the bending energy or torsional energy, and the lower the weight assigned to the corresponding neighbor by the target edge.
5. The method for predicting multi-granularity molecular properties guided by knowledge in the fields of physics and chemistry as described in claim 3, characterized in that, During the allocation of neighbor weights, if both are 1-hop neighbors, the neighbor with the larger bond angle will be assigned a lower weight to the target edge. If both are 2-hop neighbors, the neighbor with the larger dihedral angle will be assigned a lower weight to the target edge.
6. The method for predicting multi-granularity molecular properties guided by knowledge in the fields of physics and chemistry as described in claim 2, characterized in that, The calculation of the relative importance of 1-3 hop neighbors of each point based on stretching energy, bending energy, torsional energy, and nonbonded interaction energy includes: Each point updates its own representation by aggregating information from direct connecting edges and 1-3 hop neighboring points. The embedding representation of the direct connecting edges is directly passed to the target point. The bond length stretching energy between the 1-hop neighboring point and the target point is used as the aggregation weight of the 1-hop neighboring point. The bond angle bending energy and non-bond interaction energy between the 2-hop neighboring point and the target point are used as the aggregation weight of the 2-hop neighboring point. The dihedral angle torsion energy and non-bond interaction energy between the 3-hop neighboring point and the target point are used as the aggregation weight of the 3-hop neighboring point.
7. The method for predicting multi-granularity molecular properties guided by knowledge in the fields of physics and chemistry as described in claim 6, characterized in that, The weight of a 1-hop neighbor to the target point is inversely proportional to the bond length; the weight of a 2-hop neighbor to the target point is inversely proportional to the bond angle and the distance between the points; and the weight of a 3-hop neighbor to the target point is inversely proportional to the dihedral angle and the distance between the points.
8. A multi-granularity molecular property prediction system guided by knowledge from the fields of physics and chemistry, characterized in that, include: The information acquisition module is used to acquire multi-granularity information of the molecule to be predicted. The molecular property prediction module combines the physical and chemical knowledge of the molecule to be predicted, and obtains molecular characterization for molecular property prediction based on the multi-granularity information of the molecule to be predicted and the constructed molecular property prediction model. The process of constructing the molecular property prediction model includes: Treating chemical bonds within a molecule as edges in a molecular graph and atoms within a molecule as points in the molecular graph, we introduce molecular potential energy. Based on the energy of the interaction forces corresponding to each potential term in the molecular potential energy, we calculate the relative importance of each hop neighbor of each edge or point in the molecular graph and update the embedding representation of edges and points in the molecular graph. Each functional group is considered as a subgraph in the molecular graph and compressed into subgraph nodes. The 3D distance between subgraph nodes is used as the bias, and the interrelationships between subgraph nodes are considered to update the representation of the subgraph nodes. Molecular characterization is obtained based on the embedded representation of edges and points in the molecular graph, the representation of subgraph nodes, and the original molecular features.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps in the physics and chemistry knowledge-guided multi-granularity molecular property prediction method as described in any one of claims 1-7.
10. 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 program, it implements the steps in the multi-granularity molecular property prediction method guided by knowledge in the fields of physics and chemistry as described in any one of claims 1-7.