Molecular characterization model training method, molecular structure prediction method, and device
By introducing spatial angle and dihedral angle information into the molecular characterization model and combining it with the distance between atoms, the problem of inaccurate molecular characterization in existing technologies is solved, and more accurate molecular characterization and structure prediction are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BAIDU ONLINE NETWORK TECH (BEIJIBG) CO LTD
- Filing Date
- 2022-07-15
- Publication Date
- 2026-05-15
AI Technical Summary
In existing technologies, molecular characterization methods ignore the spatial distribution characteristics between atoms, resulting in inaccurate molecular characterizations.
By using the spatial angle and dihedral angle of the connecting edges of sample molecules as input, and combining the distance information between atoms, a molecular characterization model is constructed, and the accuracy of molecular characterization is improved by adjusting the model parameters.
It improves the learning ability of molecular characterization models, making molecular characterization more effective and accurate, and enabling better determination of molecular properties and structures.
Smart Images
Figure CN115274003B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to molecular characterization model training methods, molecular structure prediction methods and devices in the field of deep learning technology. Background Technology
[0002] In computational biology and computational chemistry, effective and accurate molecular characterization is crucial for understanding and predicting molecular properties, such as molecular property prediction, new drug discovery, and drug-target affinity prediction. Therefore, effectively and accurately determining the molecular characterization information of molecules has been a long-standing goal in computational biology and computational chemistry. Summary of the Invention
[0003] This disclosure provides a method for training a molecular characterization model, a method for predicting molecular structure, and an apparatus.
[0004] According to a first aspect of this disclosure, a method for training a molecular characterization model is provided, comprising:
[0005] The sample initialization edge representation of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and its neighboring edge are used as inputs to the molecular representation model to obtain the sample molecule representation information output by the molecular representation model.
[0006] Based on the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules, the parameters of the molecular characterization model are adjusted;
[0007] The sample connection edge is constructed based on the distance information between the sample atoms constituting the sample molecule. The spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the sample connection edge. The sample connection edge and its neighboring edge share a common sample atom.
[0008] According to a second aspect of this disclosure, a method for predicting molecular structures is provided, the method comprising:
[0009] The initial edge characterization of the connecting edges of the molecule to be tested, the spatial angle and dihedral angle between the connecting edges and their adjacent edges are input into the molecular characterization model to obtain the molecular characterization information of the molecule to be tested; the molecular characterization information is used to determine the molecular structure of the molecule to be tested.
[0010] The connecting edge is constructed based on the distance information between the atoms constituting the molecule to be tested. The spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the connecting edge. The connecting edge and its neighboring edge have the same atom. The molecular characterization model is trained by the molecular characterization model training method described above.
[0011] According to a third aspect of this disclosure, a molecular characterization model training device is provided, the molecular characterization model training device comprising:
[0012] The first processing module is used to take the sample initialization edge characterization of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and the adjacent edge of the sample connection edge as input to the molecular characterization model, and obtain the sample molecule characterization information output by the molecular characterization model.
[0013] The model training module is used to adjust the parameters of the molecular characterization model based on the difference between the sample molecular characterization information and the true molecular characterization information of the sample molecules.
[0014] The sample connection edge is constructed based on the distance information between the sample atoms constituting the sample molecule. The spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the sample connection edge. The sample connection edge and its neighboring edge have the same sample atom.
[0015] According to a fourth aspect of this disclosure, a molecular structure prediction device is provided, the molecular structure prediction device comprising:
[0016] The second processing module is used to input the initial edge characterization of the connecting edges included in the molecule to be tested, the spatial angle and dihedral angle between the connecting edges and their adjacent edges into the molecular characterization model to obtain the molecular characterization information of the molecule to be tested; the molecular characterization information is used to determine the molecular structure of the molecule to be tested.
[0017] The connecting edge is constructed based on the distance information between the atoms constituting the molecule to be tested. The spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the connecting edge. The connecting edge and its neighboring edge have the same atom. The molecular characterization model is trained by the molecular characterization model training method described above.
[0018] According to a fifth aspect of this disclosure, an electronic device is provided, comprising:
[0019] At least one processor; and a memory communicatively connected to said at least one processor; wherein,
[0020] The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the aforementioned molecular characterization model training method or molecular structure prediction method.
[0021] According to a sixth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the above-described molecular characterization model training method or molecular structure prediction method.
[0022] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program / instructions that, when executed by a processor, implement the molecular characterization model training method or the molecular structure prediction method described above.
[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description
[0024] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:
[0025] Figure 1 This is a schematic diagram of an optional processing flow for the molecular characterization model training method provided in this embodiment of the disclosure;
[0026] Figure 2 This is a schematic diagram of an optional processing flow for determining sample molecular characterization information provided in an embodiment of this disclosure;
[0027] Figure 3 This is a schematic diagram of an optional processing flow for determining sample molecular characterization information based on the sample edge characterization information for each spatial angle neighborhood output by the edge characterization model provided in this embodiment of the disclosure.
[0028] Figure 4 This is a detailed optional processing flow diagram of the molecular characterization model training method provided in the embodiments of this disclosure;
[0029] Figure 5 This is an optional schematic diagram of the molecular diagram provided in the embodiments of this disclosure;
[0030] Figure 6 This is an optional schematic diagram of the sample connection edge provided in the embodiments of this disclosure;
[0031] Figure 7 This is a schematic diagram of an optional processing flow for determining the spatial angle and dihedral angle between each sample connection edge and its adjacent edge in a coordinate system, provided by an embodiment of this disclosure.
[0032] Figure 8 This is a planar view obtained by projecting three spatial angular neighborhoods onto the horizontal direction of a spherical coordinate system, as provided in the embodiments of this disclosure.
[0033] Figure 9 This is a schematic diagram of an optional processing flow for dividing a spherical coordinate system into N spatial angular neighborhoods, provided in an embodiment of this disclosure;
[0034] Figure 10 This is a schematic diagram of the projection of the spatial angular neighborhood in the vertical direction of the spherical coordinate system provided in the embodiments of this disclosure;
[0035] Figure 11 This is a schematic diagram of an optional processing flow of the molecular structure prediction method provided in this embodiment of the disclosure;
[0036] Figure 12 This is a schematic diagram of the overall processing flow of the molecular structure prediction method provided in the embodiments of this disclosure;
[0037] Figure 13 This is a schematic diagram of the composition and structure of the molecular characterization model training device provided in the embodiments of this disclosure;
[0038] Figure 14 This is a schematic diagram of the composition structure of the molecular structure prediction device provided in the embodiments of this disclosure;
[0039] Figure 15 This is a block diagram of an electronic device used to implement the molecular characterization model training method or molecular structure prediction method of the embodiments of this disclosure. Detailed Implementation
[0040] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0041] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0042] In the following description, the terms “first, second, third” are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that “first, second, third” may be interchanged in a specific order or sequence where permitted, so that the embodiments of this disclosure described herein can be implemented in an order other than that illustrated or described herein.
[0043] Unless otherwise defined, 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 disclosure belongs. The terminology used herein is for the purpose of describing embodiments of this disclosure only and is not intended to be limiting of this disclosure.
[0044] Before describing the embodiments of this disclosure in detail, the relevant terms involved in this disclosure will be explained.
[0045] 1) Artificial Intelligence (AI): This refers to the theories, methods, technologies, and application systems that utilize digital computers or computer-controlled systems to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. Machine Learning (ML) is the core of AI, specifically studying how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance.
[0046] 2) The loss function is a computational function used to measure the difference between the model's predicted value and the true value. It is a non-negative real-valued function. The smaller the loss function, the better the robustness of the model.
[0047] Molecular structure is essentially a network structure composed of various types of atomic interactions. In addition to topological information, molecular structure also includes three-dimensional spatial information about the atoms, with each atom naturally distributed at different angles and distances. Three-dimensional space provides richer geometric information, such as elevation angles and dihedral angles.
[0048] In related technologies, a one-dimensional molecular SMILES format string is input into a deep neural network model, and a molecular representation is obtained by learning from the one-dimensional sequence information. Alternatively, a graph convolutional network or graph attention network is used to learn the molecular graph. However, when learning the molecular graph, most methods only consider single distance or angle information. Both of these methods for learning molecular representations ignore the spatial distribution characteristics of atoms, and fail to effectively integrate the spatial angles, dihedral angles, and distances between atoms, resulting in an inaccurate molecular representation.
[0049] Based on this, the molecular characterization model training method disclosed herein can be applied to any electronic device with molecular characterization model training capabilities, such as servers and personal computers. A schematic diagram of an optional processing flow for the molecular characterization model training method disclosed herein is shown below. Figure 1 As shown, it may include at least the following steps:
[0050] Step S101: The sample initialization edge representation of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and its neighboring edge are used as inputs to the molecular representation model to obtain the sample molecule representation information output by the molecular representation model.
[0051] In some embodiments, a molecular characterization model is trained using a sample molecule set. The sample molecule set includes multiple sample molecules, and the molecular formula of the sample molecules is represented in any form that can be recognized by electronic devices, such as in SMILES (a specification that explicitly describes molecular structure using ASCII strings). Each sample molecule in the sample training set includes annotation information, which can be the actual molecular characterization information corresponding to that sample molecule.
[0052] In some embodiments, the sample connection edge is constructed based on the distance information between the sample atoms constituting the sample molecule, and the spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the sample connection edge; two sample atoms constitute a sample connection edge, and the sample connection edge and its adjacent edge have the same sample atom.
[0053] In some embodiments, the specific implementation process for determining the molecular characterization information of a sample can be as follows: Figure 2 As shown, it includes at least the following steps:
[0054] Step S101a: Perform linear space mapping transformation on the sample initialization edge representation of the sample connection edge.
[0055] In some embodiments, a fully connected layer can be used. Perform a linear space mapping transformation on the sample initialization representation of the sample connection edges.
[0056] Step S101b: Merge the information of the two connecting edges corresponding to the dihedral angle.
[0057] In some embodiments, a fully connected layer can be used. The information of the two connecting edges corresponding to the dihedral angle is fused.
[0058] Step S101c: The sample initialization edge representation after the linear space mapping transformation and the information of the two fused connecting edges are input into the edge representation model in the molecular representation model.
[0059] In some embodiments, the edge representation model can be expressed as a function as shown in the following formula (1).
[0060]
[0061]
[0062] in, This indicates that the edge e connecting the sample in the q-th neighborhood of the l-th layer is... ij Aggregated edge representation information; Boolean function exist and The value is 1 when two adjacent edges are adjacent in a clockwise direction, otherwise... The value is 0. Used to filter neighboring edges.
[0063] and This represents a fully connected layer in a neural network that transforms edge representation information within a specific neighborhood.
[0064] in,
[0065]
[0066]
[0067] It is used to encode the two adjacent sides that form a dihedral angle and extract the contextual information of the spatial angle.
[0068] W φ,q This represents the trainable model parameters used for dihedral angle information conversion within the q-th spatial angle neighborhood learning layer. The gated recurrent unit (GRU) is used to learn the dependence of spatial sequence information perception, which can fully integrate the spatial geometric information superimposed by multiple network layers and the contextual information of neighboring edges.
[0069] Radial basis functions (RBF) have been proven to be very effective for encoding spatial information. Therefore, for encoding two-dimensional angles, we convert scalar angles into effective spatial geometric features based on RBF, as shown in the following formula (3).
[0070]
[0071] Where ⌒ represents the symbol for the K-dimensional representation vector obtained by concatenating all scalar values. Each μ k Uniform selection is made from the range of 0 to π, and As an example, if k = 6, then μ1 is 30°, μ2 is 60°, μ3 is 90°, and so on. Based on the assumption of uniform distribution, the RBF learning process can be mapped to dense vectors in fine-grained spatial neighborhood segmentation, thereby fully encoding dihedral angles.
[0072] In this embodiment of the disclosure, three types of information—spatial angle, dihedral angle, and interatomic distance—are incorporated into the molecular characterization model training process, which improves the learning ability of the molecular characterization model and enables it to effectively and accurately determine the molecular characterization of molecules.
[0073] Step S101d: Based on the sample edge representation information for the sample connection edge in each spatial angle neighborhood output by the edge representation model, determine the sample molecule representation information.
[0074] In some embodiments, an optional processing flow for determining sample molecular characterization information is based on sample edge characterization information for sample connection edges within each spatial angular neighborhood output by the edge characterization model, such as... Figure 3 As shown, it can include at least:
[0075] Step S101d1: The sample edge representation information of the sample connecting edge in each spatial angle neighborhood output by the edge representation model is spliced together to obtain the target edge representation information of the sample connecting edge.
[0076] In some embodiments, the target edge representation information can be determined by the following formula (4).
[0077]
[0078] Used to represent the sample edge representation information for sample connection edges within the Nth spatial angle neighborhood.
[0079] In step S101d2, the pooling model in the molecular characterization model performs graph pooling processing on the target edge characterization information of all sample connection edges of the sample molecule to obtain the sample molecule characterization information.
[0080] In some alternative embodiments, the pooling model may employ pooling by summing the sample edge representation information of each connected edge, as shown in the following formula (5).
[0081]
[0082] in, represents the sample edge representation information of the Lth connection edge, and h represents the sample molecule representation information.
[0083] Step S102: Adjust the parameters of the molecular characterization model based on the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules.
[0084] In some embodiments, a loss function can be constructed based on the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules; the greater the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules, the larger the value of the loss function; the smaller the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules, the larger the value of the loss function. The loss function can be a function of any form.
[0085] In practice, the form of the loss function can be determined based on the actual application scenario or task information. For example, when predicting molecular properties, the L1 loss function can be chosen, while when predicting binary drug-target interaction (DTI) properties, the cross-entropy function can be chosen as the loss function.
[0086] A detailed schematic diagram of an optional processing flow for the molecular characterization model training method provided in this disclosure is shown below. Figure 4 As shown, it may include at least the following steps:
[0087] Step S201: For each sample molecule, based on the distance information between the sample atoms constituting the sample molecule, construct sample connection edges between the sample atoms.
[0088] In some embodiments, a molecular characterization model is trained using a sample molecule set, which includes multiple sample molecules. The molecular formula of each sample molecule is represented in any form that can be recognized by an electronic device, such as in SMILES form. Each sample molecule in the training set includes annotation information, which can be the actual molecular characterization information corresponding to that sample molecule.
[0089] In some embodiments, an optional schematic diagram of the molecular diagram is shown, such as Figure 5 As shown in the diagram, various molecular diagrams reveal multiple interaction relationships between atoms due to varying distances. For example, covalent bonds exist between atoms at distances ranging from 1 to 2 angstroms; while non-covalent bonds, such as hydrophobic bonds and van der Waals forces, exist between atoms at distances greater than 2 angstroms. Therefore, different distances between atoms characterize different atomic relationships and are a crucial spatial feature of atoms.
[0090] In some embodiments, an optional implementation process for constructing sample connection edges between sample atoms based on the distance information between the sample atoms constituting the sample molecule includes: determining the relative distance between any two sample atoms; and constructing a sample connection edge between the two sample atoms if the relative distance between the two sample atoms is less than a distance threshold. Here, sample atoms are the atoms constituting the sample molecule, and the distance threshold can be flexibly set according to the actual application scenario, such as 5 angstroms or other values. When determining the relative distance between any two sample atoms, the coordinate information of the two sample atoms in the same Cartesian coordinate system can be determined first, and then the relative distance between the two sample atoms can be determined based on the coordinate information corresponding to each of the two sample atoms.
[0091] In some embodiments, an optional schematic diagram of the sample connection edge is shown, such as... Figure 6 As shown, atom a i The distances between atoms a1, a2, a3, a4, a5, a6, and a7 are all less than the distance threshold. Therefore, atoms a1, a2, a3, a4, a5, a6, and a7 are constructed respectively. i Sample connection edges between atoms a1, a2, a3, a4, a5, a6, and a7.
[0092] Step S202: Construct a coordinate system based on the sample connection edges, and determine the spatial angle and dihedral angle between each sample connection edge and its adjacent edge in the coordinate system.
[0093] In some alternative embodiments, after the sample connection edges are constructed, the angular spatial attributes of the sample connection edges are obtained; the angular spatial attributes may include: the spatial angle and dihedral angle between the sample connection edge and its adjacent edge.
[0094] In some alternative embodiments, with Figure 6 Taking the spherical coordinate system shown as an example, the sample connection edge is formed by atom a i and atom a j The edge formed, a i The starting point of the sample connection edge is atom a. i and atom a j Establish a spherical coordinate system with the connecting edges of the formed samples as the Z-axis. In the spherical coordinate system, the axis perpendicular to atom a... i and atom a j The largest plane forming the sample connection edges is the reference plane. Based on Figure 6 This allows us to determine the spatial angle and dihedral angle between a sample connection edge and any adjacent edge of another sample connection edge; as an example, the sample connection edge connects to atoms a2 and a... i The spatial angle between the adjacent edges formed is θ2, and the sample connecting edge is connected to atom a1 and atom a.i The spatial angle between the adjacent edges formed is θ1, and the sample connecting edge is connected to atom a3 and atom a. i The spatial angle between the adjacent sides formed is θ3. Atom a2 and atom a i The projection of the adjacent sides formed onto the reference plane, and the relationship between atoms a1 and a. i The angle between the projections of the adjacent sides onto the reference plane It is a dihedral angle.
[0095] Based on the above optional embodiments, an optional processing procedure for determining the spatial angle and dihedral angle between each sample connection edge and its adjacent edge in the coordinate system is as follows: Figure 7 As shown, it may include:
[0096] Step S202a: Construct a spherical coordinate system for each sample molecule.
[0097] In some embodiments, a spherical coordinate system can be constructed with the sample connection edge formed by any two atoms in the sample molecule as the Z-axis of the spherical coordinate system.
[0098] Step S202b: Determine the horizontal projection of the sample connection edge in the spherical coordinate system to obtain a planar view of each sample atom in the sample molecule.
[0099] In some embodiments, the spherical coordinate system is first divided into N spatial angle neighborhoods based on the sample connection edges; then the horizontal projection of each spatial angle neighborhood in the spherical coordinate system is determined; and finally the horizontal projection of the sample connection edges in the spherical coordinate system is determined.
[0100] In some embodiments, with Figure 6 Taking the spherical coordinate system shown as an example, the spherical coordinate system includes three spatial angular neighborhoods, i.e., N=3; projecting the three spatial angular neighborhoods onto the horizontal direction of the spherical coordinate system, we obtain the following... Figure 8 The plan view shown is based on Figure 8 The planar view shown can determine the planar view of each sample atom in the sample molecule. Thus, by dividing the spherical coordinate system into multiple spatial angular neighborhoods, the representational information of the connecting edges carries spatial angular information.
[0101] In some alternative embodiments, an optional processing flow involves dividing the spherical coordinate system into N spatial angular neighborhoods, such as... Figure 9 As shown, it includes at least the following steps:
[0102] Step S202b1: Based on the sample connection edges, divide the spherical coordinate system into X candidate spatial angle neighborhoods.
[0103] In some embodiments, the sample initialization edge features of each sample atom's sample connection edge can be determined based on the sample initialization atom features of the two sample atoms corresponding to the sample connection edge and the distance representation vector between the two sample atoms. In a specific implementation, this can be achieved by aggregating the atoms a at both ends of the sample connection edge. i and a j Get sample connection edge e ij The sample initialization edge representation is used, and angle information is embedded during the aggregation process; the sample connection edge e ij The sample initialization edge representation is shown in the following formula (6):
[0104] e ij =σ(W a ·[a i ||a j ||d ij (6)
[0105] Here, || represents the concatenation operation, and the ReLU function can be used as the activation function σ, W. a It is the trainable model parameter matrix, a i and a j d represents the initial atomic features of the input. ij d represents the distance representation vector; ij The distance r can be encoded using methods such as distance discretization and radial basis functions. The encoding process for the distance r can employ existing techniques, which will not be elaborated upon here. Step b1 can obtain the sample initialization edge representation of the sample edges and can fuse the important spatial distance attributes of each sample atom.
[0106] In some embodiments, the spherical coordinate system corresponding to the molecular structure is divided into X candidate angular neighborhoods based on the spatial angle between the sample connection edge and each neighboring edge; as an example, X = 3.
[0107] Step S202b2: Determine the set of neighboring edges within the neighborhood of each candidate spatial angle.
[0108] In some embodiments, based on Figure 8 The planar view shown indicates that each neighbor edge can be located within the planar view by distance r and spatial angle θ. The neighborhood divider can determine the number of each spatial angle neighborhood based on θ. The formula for determining the number of the spatial angle neighborhood is shown in the following formula (7):
[0109]
[0110] Among them, Ind ki This represents the index of the spatial angular neighborhood, where Q is the number of spatial angular neighborhoods; e ijIndicates the sample connection edge, e ki Indicates the adjacent edge of the sample connection edge. The integer symbol θ represents the rounding off. kij ∈[0,360°] represents edge e ij With e ki The angle between them.
[0111] Based on the spatial angular neighborhood of each adjacent edge, the subset of adjacent edges located in the q-th spatial angular neighborhood can be determined as shown in the following formula (8):
[0112]
[0113] Among them, Ind ki =q indicates that the spatial angular neighborhood is numbered q. This represents the set of neighboring edges of the sample connecting edges within the spatial angular neighborhood.
[0114] Step S202b3: Based on the projection of each spatial angle neighborhood in the horizontal direction of the spherical coordinate system and the set of adjacent edges in each candidate spatial angle neighborhood, update the X candidate spatial angle neighborhoods into N spatial angle neighborhoods.
[0115] In some embodiments, given a subset of adjacent edges already partitioned into spatial angular neighborhoods, a neighborhood partitioner can be used to repartition all adjacent edges. In a specific implementation, a spherical cone is formed on the horizontal projection of each candidate spatial angular neighborhood, and each spherical cone contains multiple adjacent edges within its local space. N spatial angular neighborhoods are determined based on the set of adjacent edges within each candidate spatial supervisory neighborhood. Therefore, in this embodiment, the spatial angular neighborhood carries information about the spatial angle θ.
[0116] In this embodiment of the disclosure, by dividing the spatial angular neighborhood, the problem of model learning being hindered by the dense distribution of neighboring edges in space can be alleviated.
[0117] In some embodiments, N and X can be the same value. When updating X candidate spatial angle neighborhoods to N spatial angle neighborhoods, it can be an update of the number of spatial angle neighborhoods, such as updating from 3 candidate spatial angle neighborhoods to 4 spatial angle neighborhoods. Alternatively, it can be that the connecting edges within each spatial angle neighborhood change, such as updating from 3 candidate spatial angle neighborhoods to 3 spatial angle neighborhoods, where at least one neighboring edge within a spatial angle neighborhood is not exactly the same as the neighboring edge within the corresponding candidate spatial angle neighborhood.
[0118] Step S202c: Based on the planar view of each sample atom, determine the spatial angle between each sample connection edge and its adjacent edge.
[0119] Step S202d: Determine the dihedral angle based on the spatial angle between each sample connection edge and its adjacent edge.
[0120] In some embodiments, the specific implementation process of determining the dihedral angle may include: determining N two-dimensional neighborhood planes based on the projection of each spatial angle neighborhood in the vertical direction of the spherical coordinate system; and determining the included angle between any two sides in each two-dimensional neighborhood plane as the dihedral angle between the sample connecting edge and the two adjacent sides forming the two faces respectively.
[0121] In some embodiments, Figure 6 The projection of the spatial angular neighborhood onto the perpendicular direction in the spherical coordinate system is shown below. Figure 10 As shown, there are three two-dimensional neighborhood planes; two-dimensional neighborhood planes A1 and A2 each have two edges, and A1 and A2 each contain two dihedral angles; two-dimensional neighborhood plane A3 has three edges, including three dihedral angles. The sum of the dihedral angles in any neighborhood plane is 360°.
[0122] Step S203: The sample initialization edge representation of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and its neighboring edge are used as inputs to the molecular representation model to obtain the sample molecule representation information output by the molecular representation model.
[0123] Step S204: Adjust the parameters of the molecular characterization model based on the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules.
[0124] In some embodiments, the specific implementation process of steps S203 to S204 is the same as that of steps S101 to S102, and will not be described again here.
[0125] This disclosure also provides a molecular structure prediction method. The molecular structure prediction method provided by this disclosure can be applied to any electronic device with molecular structure prediction capabilities. The execution subject of the molecular characterization model training method provided by this disclosure and the molecular structure prediction method provided by this disclosure can be the same or different.
[0126] An optional processing flow of the molecular structure prediction method provided in this disclosure embodiment may include at least:
[0127] The initial edge representation of the connecting edge, the spatial angle and dihedral angle between the connecting edge and its adjacent edge are input into the molecular characterization model to obtain the molecular characterization information of the molecule to be tested; the molecular characterization information is used to determine the molecular structure of the molecule to be tested.
[0128] In some embodiments, the connecting edge is constructed based on the distance information between the atoms constituting the molecule under test, and the spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the connecting edge; two atoms of the molecule under test constitute a connecting edge, and the connecting edge and its adjacent edge have the same atom.
[0129] In some embodiments, the molecular characterization model is based on Figure 1 The neural network model trained by the molecular characterization model training method is described.
[0130] A detailed optional processing flow diagram of the molecular structure prediction method provided in this disclosure embodiment is shown below, such as... Figure 11 As shown, it can include at least:
[0131] Step S401: Based on the distance information between the atoms of the molecule to be tested, construct the connection edges between the atoms.
[0132] In some embodiments, the process of constructing the connection edges between the atoms based on the distance information between the atoms of the molecule to be tested can be the same as the process of constructing the sample connection edges between the sample atoms based on the distance information between the sample atoms constituting the sample molecule in step S201, and will not be described again here.
[0133] Step S402: Construct a coordinate system based on the connecting edges, and determine the spatial angle and dihedral angle between each connecting edge and its adjacent edge in the coordinate system.
[0134] In some embodiments, the process of determining the spatial angle and dihedral angle between each connecting edge and its adjacent edge can be the same as the process of determining the spatial angle and dihedral angle between sample connecting edges and their adjacent edges in step S202, and will not be described again here.
[0135] Step S403: Input the initial edge representation of the connecting edge, the spatial angle and dihedral angle between the connecting edge and its neighboring edge into the molecular representation model to obtain the molecular representation information of the molecule to be tested; the molecular representation information is used to determine the molecular structure of the molecule to be tested.
[0136] In some embodiments, the molecular characterization model is based on Figure 1 The neural network model trained using the molecular characterization model training method shown is an example.
[0137] based on Figures 1 to 11 As shown in the schematic diagram of the overall processing flow of the molecular structure prediction method provided in this embodiment, as follows: Figure 12As shown, after determining the molecular graph to be tested, connecting edges are constructed based on the molecular graph to be tested; the schematic diagram of the connecting edges is input into the pre-trained molecular representation model, and the node-edge transformation layer in the molecular representation model transforms the input connecting edge schematic diagram. After being processed by two edge-edge geometry perception layers and an edge-node transformation layer, the molecular graph is processed by molecular graph pooling to obtain molecular representation information; downstream tasks such as molecular property prediction are performed based on the molecular representation information.
[0138] This disclosure also provides a molecular characterization model training device, the composition and structure of which are as follows: Figure 13 As shown, it includes:
[0139] The first processing module 501 is used to take the sample initialization edge characterization of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and the adjacent edge of the sample connection edge as input to the molecular characterization model, and obtain the sample molecule characterization information output by the molecular characterization model.
[0140] The model training module 502 is used to adjust the parameters of the molecular characterization model based on the difference between the sample molecular characterization information and the true molecular characterization information of the sample molecules.
[0141] The sample connection edge is constructed based on the distance information between the sample atoms constituting the sample molecule. The spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the sample connection edge. The sample connection edge and its neighboring edge share a common sample atom.
[0142] In some alternative embodiments, the molecular characterization model training device further includes: a first building module ( Figure 13 (not shown in the image), used to determine the relative distance between any two sample atoms in the sample molecule;
[0143] If the relative distance between two sample atoms is less than a distance threshold, a sample connection edge is constructed between the two sample atoms.
[0144] In some optional embodiments, the molecular characterization model training device further includes: a first spatial angle information determination module ( Figure 13 (Not shown in the image), used to construct a spherical coordinate system for each of the sample molecules;
[0145] Determine the horizontal projection of the sample connection edge in the spherical coordinate system to obtain a planar view of each sample atom in the sample molecule;
[0146] Based on the planar view of each sample atom, determine the spatial angle between each sample connection edge and its adjacent edge;
[0147] The dihedral angle is determined based on the spatial angle between each sample connection edge and its adjacent edge.
[0148] In some optional embodiments, the first spatial angle information determination module is used to divide the spherical coordinate system into N spatial angle neighborhoods based on the sample connection edges, where N is greater than or equal to 2;
[0149] Based on the projection of each spatial angular neighborhood into the spherical coordinate system, N two-dimensional neighborhood planes are determined.
[0150] The included angle between any two edges in each two-dimensional neighborhood plane is defined as the dihedral angle between the sample connecting edge and the two adjacent edges forming the two faces.
[0151] In some optional embodiments, the first spatial angle information determination module is used to divide the spherical coordinate system into X candidate spatial angle neighborhoods based on the sample connection edges;
[0152] Determine the set of neighboring edges within each candidate spatial angular neighborhood;
[0153] Based on the projection of each spatial angle neighborhood into the horizontal direction of the spherical coordinate system and the set of adjacent edges within each candidate spatial angle neighborhood, the X candidate spatial angle neighborhoods are updated into N spatial angle neighborhoods.
[0154] In some alternative embodiments, the first construction module is further configured to determine the sample initialization edge features of the sample connection edge of each sample atom based on the sample initialization atom features of the two sample atoms corresponding to the sample connection edge and the distance representation vector between the two sample atoms.
[0155] In some optional embodiments, the first processing module 501 is used to perform a linear space mapping transformation on the sample initialization edge representation of the sample connection edge;
[0156] The information of the two connecting edges corresponding to the dihedral angle is fused;
[0157] The sample initialization edge representation after the linear space mapping transformation and the information of the two fused connecting edges are input into the edge representation model in the molecular representation model;
[0158] Based on the sample edge representation information for the sample connection edge in each spatial angle neighborhood output by the edge representation model, the sample molecule representation information is determined.
[0159] In some optional embodiments, the first processing module 501 is used to splice the sample edge representation information for the sample connecting edge in each spatial angle neighborhood output by the edge representation model to obtain the target edge representation information of the sample connecting edge.
[0160] The pooling model in the molecular characterization model performs graph pooling processing on the target edge characterization information of all sample connection edges of the sample molecule to obtain the characterization information of the sample molecule.
[0161] This disclosure also provides a molecular structure prediction device, the composition of which is as follows: Figure 14 As shown, it includes:
[0162] The second processing module 601 is used to input the initial edge characterization of the connecting edge of the molecule to be tested, the spatial angle and dihedral angle between the connecting edge and its adjacent edge into the molecular characterization model to obtain the molecular characterization information of the molecule to be tested; the molecular characterization information is used to determine the molecular structure of the molecule to be tested.
[0163] The connecting edge is constructed based on the distance information between the atoms constituting the molecule under test. The spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the connecting edge. The connecting edge and its neighboring edge have the same atom. The molecular characterization model is trained by the above-mentioned molecular characterization model training method.
[0164] In some embodiments, the molecular structure prediction device further includes a second building module ( Figure 14 (not shown in the image), used to construct the connection edges between the atoms based on the distance information between the atoms of the molecule to be tested;
[0165] Second spatial angle information determination module ( Figure 14 (not shown in the image), used to construct a coordinate system based on the connecting edge, and to determine the spatial angle and dihedral angle between each connecting edge and its adjacent edge in the coordinate system.
[0166] It should be noted that the acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0167] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.
[0168] Figure 15A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. In some alternative embodiments, the electronic device 800 may be a terminal device or a server. In some alternative embodiments, the electronic device 800 may implement the molecular characterization model training method or molecular structure prediction method provided in the embodiments of this application by running a computer program. For example, the computer program may be a native program or software module in an operating system; it may be a native application (APP), i.e., a program that needs to be installed in the operating system to run; it may be a small program, i.e., a program that only needs to be downloaded to a browser environment to run; or it may be a small program that can be embedded in any APP. In summary, the above-described computer program may be any form of application, module, or plugin.
[0169] In practical applications, electronic device 800 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 functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. Electronic device 800 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smart TV, smartwatch, etc., but is not limited to these.
[0170] Electronic devices are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, in-vehicle terminals, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0171] like Figure 15As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.
[0172] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0173] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as molecular characterization model training methods or molecular structure prediction methods. For example, in some alternative embodiments, the molecular characterization model training method or molecular structure prediction method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some alternative embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the molecular characterization model training method or molecular structure prediction method described above can be performed. Alternatively, in other embodiments, the computing unit 801 may be configured by any other suitable means (e.g., by means of firmware) as a molecular characterization model training method or a molecular structure prediction method.
[0174] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0175] The program code used to implement the molecular characterization model training method or molecular structure prediction method of this disclosure can be written in any combination of one or more programming languages. This program code can be provided to the processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.
[0176] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0177] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0178] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.
[0179] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.
[0180] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0181] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A method for training a molecular characterization model, comprising: Construct a spherical coordinate system for the sample molecules; Determine the horizontal projection of the sample connection edge in the spherical coordinate system to obtain a planar view of each sample atom in the sample molecule; Based on the planar view of each sample atom, determine the spatial angle between each sample connection edge and its adjacent edge; The dihedral angle is determined based on the spatial angle between each sample connection edge and its adjacent edge; The sample initialization edge representation of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and its neighboring edge are used as inputs to the molecular representation model to obtain the sample molecule representation information output by the molecular representation model. Based on the difference between the sample molecule characterization information and the true molecular characterization information of the sample molecules, the parameters of the molecular characterization model are adjusted; Wherein, the sample connection edge is constructed based on the distance information between each sample atom constituting the sample molecule, the spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the sample connection edge, and the sample connection edge and its adjacent edge share a common sample atom; The step of determining the dihedral angle based on the spatial angle between each sample connecting edge and its adjacent edge includes: dividing the spherical coordinate system into N spatial angle neighborhoods based on the sample connecting edge, where N is greater than or equal to 2; determining N two-dimensional neighborhood planes based on the projection of each spatial angle neighborhood into the vertical direction of the spherical coordinate system; and determining the included angle between any two edges in each two-dimensional neighborhood plane as the dihedral angle between the sample connecting edge and the two adjacent edges forming the two faces.
2. The method according to claim 1, wherein, The method further includes: Determine the relative distance between any two sample atoms in the sample molecule; If the relative distance between the two sample atoms is less than a distance threshold, a sample connection edge is constructed between the two sample atoms.
3. The method according to claim 1, wherein, The step of dividing the spherical coordinate system into N spatial angular neighborhoods based on the sample connection edges includes: Based on the sample connection edges, the spherical coordinate system is divided into X candidate spatial angle neighborhoods; Determine the set of neighboring edges within each candidate spatial angular neighborhood; Based on the projection of each spatial angle neighborhood into the horizontal direction of the spherical coordinate system and the set of adjacent edges within each candidate spatial angle neighborhood, the X candidate spatial angle neighborhoods are updated into N spatial angle neighborhoods.
4. The method according to claim 1, wherein, The method further includes: Based on the sample initialization atom features of the two sample atoms corresponding to the sample connection edge, and the distance representation vector between the two sample atoms, the sample initialization edge features of the sample connection edge of each sample atom are determined.
5. The method according to claim 1, wherein, The process of using the sample initialization edge representation of the sample connecting edge, the spatial angle and dihedral angle between the sample connecting edge and its adjacent edge as inputs to the molecular characterization model, and obtaining the sample molecular characterization information output by the molecular characterization model includes: Perform a linear space mapping transformation on the sample initialization edge representation of the sample connection edge; The information of the two connecting edges corresponding to the dihedral angle is fused; The sample initialization edge representation after the linear space mapping transformation and the information of the two fused connecting edges are input into the edge representation model in the molecular representation model; Based on the sample edge representation information for the sample connection edge in each spatial angle neighborhood output by the edge representation model, the sample molecule representation information is determined.
6. The method according to claim 1, wherein, The determination of the sample molecule representation information based on the sample edge representation information for each spatial angle neighborhood output by the edge representation model includes: The sample edge representation information for the sample connecting edge in each spatial angle neighborhood output by the edge representation model is concatenated to obtain the target edge representation information of the sample connecting edge. The pooling model in the molecular characterization model performs graph pooling processing on the target edge characterization information of all sample connection edges of the sample molecule to obtain the characterization information of the sample molecule.
7. A method for predicting molecular structure, the method comprising: The initial edge characterization of the connecting edges of the molecule to be tested, the spatial angle and dihedral angle between the connecting edges and their adjacent edges are input into the molecular characterization model to obtain the molecular characterization information of the molecule to be tested; the molecular characterization information is used to determine the molecular structure of the molecule to be tested. Wherein, the connecting edge is constructed based on the distance information between the atoms constituting the molecule to be tested, the spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the connecting edge, the connecting edge and its neighboring edge have the same atom, and the molecular characterization model is trained by the molecular characterization model training method as described in any one of claims 1 to 6.
8. A molecular characterization model training device, the molecular characterization model training device comprising: The first spatial angle information determination module is used to construct a spherical coordinate system for sample molecules; Determine the horizontal projection of the sample connection edge in the spherical coordinate system to obtain a planar view of each sample atom in the sample molecule; Based on the planar view of each sample atom, determine the spatial angle between each sample connecting edge and its adjacent edge; based on the spatial angle between each sample connecting edge and its adjacent edge, determine the dihedral angle; The first processing module is used to take the sample initialization edge characterization of the sample connection edge corresponding to the sample molecule, the spatial angle and dihedral angle between the sample connection edge and the adjacent edge of the sample connection edge as input to the molecular characterization model, and obtain the sample molecule characterization information output by the molecular characterization model. The model training module is used to adjust the parameters of the molecular characterization model based on the difference between the sample molecular characterization information and the true molecular characterization information of the sample molecules. Wherein, the sample connection edge is constructed based on the distance information between each sample atom constituting the sample molecule, the spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the sample connection edge, and the sample connection edge and its adjacent edge have the same sample atom; The first spatial angle information determination module is further configured to divide the spherical coordinate system into N spatial angle neighborhoods based on the sample connecting edge, where N is greater than or equal to 2; determine N two-dimensional neighborhood planes based on the projection of each spatial angle neighborhood in the vertical direction of the spherical coordinate system; and determine the included angle between any two edges in each two-dimensional neighborhood plane as the dihedral angle between the sample connecting edge and the two adjacent edges forming the two faces.
9. A molecular structure prediction device, the molecular structure prediction device comprising: The second processing module is used to input the initial edge characterization of the connecting edges included in the molecule to be tested, the spatial angle and dihedral angle between the connecting edges and their adjacent edges into the molecular characterization model to obtain the molecular characterization information of the molecule to be tested; the molecular characterization information is used to determine the molecular structure of the molecule to be tested. Wherein, the connecting edge is constructed based on the distance information between the atoms constituting the molecule to be tested, the spatial angle and the dihedral angle are angle information in the coordinate system constructed based on the connecting edge, the connecting edge and its neighboring edge have the same atom, and the molecular characterization model is trained by the molecular characterization model training method as described in any one of claims 1 to 6.
10. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6, or to perform the method of claim 7.
11. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 6, or to perform the method according to claim 7.
12. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the method of any one of claims 1 to 6, or implement the method of claim 7.