CN114429801A - Data processing method, training method, recognition method, device, equipment and medium - Google Patents

Data processing method, training method, recognition method, device, equipment and medium Download PDF

Info

Publication number
CN114429801A
CN114429801A CN202210097241.7A CN202210097241A CN114429801A CN 114429801 A CN114429801 A CN 114429801A CN 202210097241 A CN202210097241 A CN 202210097241A CN 114429801 A CN114429801 A CN 114429801A
Authority
CN
China
Prior art keywords
layer
target
map
feature vector
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210097241.7A
Other languages
Chinese (zh)
Inventor
刘荔行
何东龙
方晓敏
王凡
何径舟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Baidu Netcom Science and Technology Co Ltd
Original Assignee
Beijing Baidu Netcom Science and Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Baidu Netcom Science and Technology Co Ltd filed Critical Beijing Baidu Netcom Science and Technology Co Ltd
Priority to CN202210097241.7A priority Critical patent/CN114429801A/en
Publication of CN114429801A publication Critical patent/CN114429801A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16CCOMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70Machine learning, data mining or chemometrics
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16CCOMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/30Prediction of properties of chemical compounds, compositions or mixtures
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16CCOMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/80Data visualisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Artificial Intelligence (AREA)
  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The present disclosure provides a data processing method, a training method for a feature extraction model, a molecular attribute recognition method, an apparatus, an electronic device, a storage medium, and a program product, and relates to the technical field of data processing, in particular to the technical field of deep learning and biological computation. The specific implementation scheme is as follows: determining a target map by utilizing the molecular structure information of the molecules to be processed, wherein the target map is used for representing a plurality of molecular space structure information with different dimensions; and determining a spatial structural feature representation of the molecule to be processed based on the target profile.

Description

Data processing method, training method, recognition method, device, equipment and medium
Technical Field
The present disclosure relates to the field of data processing technologies, and in particular, to the field of deep learning and biometric computing technologies, and in particular, to a data processing method, a training method for a feature extraction model, a molecular attribute recognition method, an apparatus, an electronic device, a storage medium, and a program product.
Background
With the continuous development of computer technology, the research phase of chemical materials and medicines by using experimental instruments has gradually progressed to the research phase of using novel machine learning technology. The novel machine learning technology applied to molecular and material science combines the statistical viewpoint with the traditional chemical technology and has research potential.
Disclosure of Invention
The present disclosure provides a data processing method, a training method of a feature extraction model, a molecular attribute recognition method, an apparatus, an electronic device, a storage medium, and a program product.
According to an aspect of the present disclosure, there is provided a data processing method including: determining a target map by utilizing molecular structure information of molecules to be processed, wherein the target map is used for representing a plurality of molecular space structure information with different dimensions; and determining a spatial structural feature representation of the molecule to be processed based on the target profile.
According to another aspect of the present disclosure, there is provided a training method of a feature extraction model, including: determining a target sample map by using the molecular structure information of the sample molecules, wherein the target sample map is used for representing a plurality of molecular space structure information with different dimensions; processing the target sample map by using a feature extraction model, and determining sample space structure feature representation of the sample molecules; determining attribute prediction information of the sample molecules based on the sample spatial structure feature representation; and training the feature extraction model by using the attribute prediction information of the sample molecules and the labels of the sample molecules to obtain the trained feature extraction model, wherein the labels are used for representing the attribute information of the sample molecules.
According to another aspect of the present disclosure, there is provided a molecular property identification method, including: processing a molecule to be identified by using the method disclosed by the disclosure, and determining a spatial structure characteristic representation of the molecule to be identified; and determining target attribute information of the molecules to be identified based on the spatial structure feature representation.
According to another aspect of the present disclosure, there is provided a data processing apparatus including: the device comprises a first determination module, a second determination module and a third determination module, wherein the first determination module is used for determining a target map by utilizing molecular structure information of molecules to be processed, and the target map is used for representing a plurality of molecular space structure information with different dimensions; and a second determination module for determining a spatial structure characteristic representation of the molecule to be processed based on the target profile.
According to another aspect of the present disclosure, there is provided a training apparatus for a feature extraction model, including: the sample map determining module is used for determining a target sample map by utilizing the molecular structure information of sample molecules, wherein the target sample map is used for representing a plurality of molecular space structure information with different dimensions; the sample extraction module is used for processing the target sample map by using a feature extraction model and determining the sample space structure feature representation of the sample molecules; a prediction module for determining attribute prediction information of the sample molecules based on the sample spatial structure feature representation; and the training module is used for training the feature extraction model by utilizing the attribute prediction information of the sample molecules and the labels of the sample molecules to obtain the trained feature extraction model, wherein the labels are used for representing the attribute information of the sample molecules.
According to another aspect of the present disclosure, there is provided a molecular property identification apparatus including: a feature representation determining module, configured to process a molecule to be identified by using the data processing apparatus of the present disclosure, and determine a spatial structure feature representation of the molecule to be identified; and the identification module is used for determining target attribute information of the molecules to be identified based on the spatial structure feature representation.
According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method according to the present disclosure.
According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform a method as disclosed herein.
According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements a method as disclosed herein.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.
Drawings
The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:
FIG. 1 schematically illustrates an exemplary system architecture to which the molecular property identification methods and apparatus may be applied, according to embodiments of the present disclosure;
FIG. 2 schematically shows a flow chart of a data processing method according to an embodiment of the present disclosure;
figure 3A schematically shows a three-dimensional structural schematic of a methylamine molecule according to an embodiment of the disclosure;
figure 3B schematically shows a schematic of an atom-chemical bond target sub-map of a methylamine molecule according to an embodiment of the present disclosure;
figure 3C schematically shows a schematic diagram of a chemical bond-angle target sub-map of a methylamine molecule according to an embodiment of the present disclosure;
figure 3D schematically shows a schematic of a bond angle-dihedral angle target sub-map of methylamine molecules according to an embodiment of the present disclosure;
FIG. 4 schematically illustrates a network architecture diagram of a feature extraction model according to an embodiment of the disclosure;
FIG. 5 schematically illustrates a 2-butene molecular structure schematic according to an embodiment of the present disclosure;
FIG. 6 schematically shows a flow diagram of a method of training a feature extraction model according to an embodiment of the present disclosure;
FIG. 7 schematically illustrates a flow chart of a molecular property identification method according to an embodiment of the present disclosure;
FIG. 8 schematically shows a block diagram of a data processing apparatus according to an embodiment of the present disclosure;
FIG. 9 schematically shows a block diagram of a training apparatus for a feature extraction model according to an embodiment of the present disclosure;
FIG. 10 schematically illustrates a block diagram of a molecular property identification apparatus according to an embodiment of the present disclosure; and
fig. 11 schematically shows a block diagram of an electronic device adapted to implement a content processing method according to an embodiment of the present disclosure.
Detailed Description
Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The present disclosure provides a data processing method, a training method of a feature extraction model, a molecular attribute recognition method, an apparatus, an electronic device, a storage medium, and a program product.
According to an embodiment of the present disclosure, there is provided a data processing method including: determining a target map by utilizing the molecular structure information of the molecules to be processed, wherein the target map is used for representing a plurality of molecular space structure information with different dimensions; and determining a spatial structural feature representation of the molecule to be processed based on the target profile.
According to an embodiment of the present disclosure, there is provided a training method of a feature extraction model, including: determining a target sample map by using the molecular structure information of the sample molecules, wherein the target sample map is used for representing a plurality of molecular space structure information with different dimensions; processing a target sample map by using a feature extraction model, and determining sample space structure feature representation of sample molecules; determining attribute prediction information of sample molecules based on the spatial structure feature representation of the sample; and training the feature extraction model by using the attribute prediction information of the sample molecules and the labels of the sample molecules to obtain the trained feature extraction model, wherein the labels are used for representing the attribute information of the sample molecules.
According to an embodiment of the present disclosure, there is provided a molecular attribute identification method, including: processing the molecules to be identified by using the method disclosed by the invention, and determining the spatial structure characteristic representation of the molecules to be identified; and determining target attribute information of the molecules to be identified based on the spatial structure feature representation.
In the technical scheme of the disclosure, the collection, storage, use, processing, transmission, provision, disclosure and other processing of the personal information of the related user are all in accordance with the regulations of related laws and regulations and do not violate the good customs of the public order.
Fig. 1 schematically illustrates an exemplary system architecture to which the molecular property identification method and apparatus may be applied, according to an embodiment of the present disclosure.
It should be noted that fig. 1 is only an example of a system architecture to which the embodiments of the present disclosure may be applied to help those skilled in the art understand the technical content of the present disclosure, and does not mean that the embodiments of the present disclosure may not be applied to other devices, systems, environments or scenarios. For example, in another embodiment, an exemplary system architecture to which the method and apparatus for identifying a molecular attribute may be applied may include a terminal device, but the terminal device may implement the method and apparatus for identifying a molecular attribute provided in the embodiments of the present disclosure without interacting with a server.
As shown in fig. 1, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 serves as a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired and/or wireless communication links, and so forth.
The user may use the terminal devices 101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. The terminal devices 101, 102, 103 may have installed thereon various communication client applications, such as a knowledge reading application, a web browser application, a search application, an instant messaging tool, a mailbox client, and/or social platform software, etc. (by way of example only).
The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.
The server 105 may be a server providing various services, such as a background management server (for example only) providing support for content browsed by the user using the terminal devices 101, 102, 103. The background management server may analyze and perform other processing on the received data such as the user request, and feed back a processing result (e.g., a webpage, information, or data obtained or generated according to the user request) to the terminal device.
It should be noted that the molecular property identification method provided by the embodiment of the present disclosure may be generally executed by the terminal device 101, 102, or 103. Accordingly, the molecular property recognition device provided by the embodiment of the present disclosure may also be disposed in the terminal device 101, 102, or 103.
Alternatively, the molecular property identification method provided by the embodiments of the present disclosure may also be generally performed by the server 105. Accordingly, the molecular property recognition device provided by the embodiment of the present disclosure may be generally disposed in the server 105. The molecular property identification method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the molecular property identification apparatus provided by the embodiment of the present disclosure may also be disposed in a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.
For example, when a user wants to obtain target attribute information of a drug molecule, such as drug toxicity, drug ligand, affinity of a protein receptor, and the like, by using a molecular expression of the drug molecule, the terminal devices 101, 102, 103 may obtain the molecular expression input by the user, then send the obtained molecular expression to the server 105, and the server 105 analyzes the molecular expression to determine molecular structure information of the drug molecule; determining a target profile based on the molecular structure information; determining a spatial structure characteristic representation of the drug molecule based on the target map; target attribute information of the drug molecules is determined based on the spatial structural feature representation. Or by a server or server cluster capable of communicating with the terminal devices 101, 102, 103 and/or the server 105, to analyze the molecular expressions of the drug molecules and finally determine the target property information of the drug molecules.
It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
It should be noted that the sequence numbers of the respective operations in the following methods are merely used as representations of the operations for description, and should not be construed as representing the execution order of the respective operations. The method need not be performed in the exact order shown, unless explicitly stated.
Fig. 2 schematically shows a flow chart of a data processing method according to an embodiment of the present disclosure.
As shown in fig. 2, the method includes operations S210 to S220.
In operation S210, a target map is determined using the molecular structure information of the molecules to be processed, wherein the target map is used to represent a plurality of molecular spatial structure information of different dimensions.
In operation S220, a spatial structural feature representation of the molecule to be processed is determined based on the target profile.
According to an embodiment of the present disclosure, the molecular structure information may include molecular property information, but is not limited thereto, and may also include molecular spatial structure information. The molecular structure information may include one or more of atoms, chemical bonds, bond lengths, bond angles, dihedral angles.
According to embodiments of the present disclosure, the target graph, may also be referred to as a target topology. The target graph may include vertices that may be used to represent one object and edges that may be used to represent associations between multiple objects. For example, the atoms of the molecule to be treated correspond to vertices in the target map, and the chemical bonds between the atoms may correspond to edges in the target map, i.e., atom-chemical bond target sub-maps. But is not limited thereto. The molecular spatial information may further comprise bond length, bond angle, dihedral angle, etc., and the target pattern may comprise a plurality of target sub-patterns, for example, one or more of a chemical bond-bond angle target sub-pattern, a bond angle-dihedral angle target sub-pattern, in addition to the atom-chemical bond target sub-pattern.
According to an embodiment of the present disclosure, the spatial structure feature representation may be referred to as a spatial structure feature vector, a spatial structure feature sequence, or the like. Spatial structural feature representation can be used to characterize spatial structural information and properties in the molecule to be processed, and further, prediction of properties of the molecule to be processed, such as drug toxicity, drug ligand, affinity of protein receptor, etc., can be predicted based on the spatial structural feature representation using methods such as deep learning.
According to embodiments of the present disclosure, a target profile may be used to characterize multiple molecular spatial structure information of different dimensions. The plurality of molecular spatial structure information may be constructed based on a variety of information among the molecular structure information including atoms, chemical bonds, bond lengths, bond angles, dihedral angles, and the like. Therefore, the information carried in the target map can be sufficient and complete, the generated information represented by the spatial structure characteristics can represent more fine spatial structure information of the molecules to be processed, and the problem that the spatial structure information of the molecules to be processed is not sufficiently captured by the spatial structure characteristic representation due to the fact that the information carried by the target map is incomplete is solved.
According to an embodiment of the present disclosure, before determining the target profile using the molecular structure information of the molecule to be processed in operation S210, the data processing method may further perform the following operation.
For example, the molecular expression of the molecule to be treated is determined. Based on the molecular expression, molecular structure information of the molecule to be processed is determined.
According to embodiments of the present disclosure, the Molecular expressions may include Simplified Molecular Input Line Entry Specification (SMILES), but are not limited thereto and may include other expressions of Specification that explicitly describe the Molecular structure in ASCII strings.
According to the embodiment of the present disclosure, the molecular structure information of the molecule to be processed may be obtained based on a molecular expression in a molecular dynamics simulation or experimental measurement and calculation manner, and is not described herein again.
Fig. 3A schematically shows a three-dimensional structural schematic of a methylamine molecule according to an embodiment of the disclosure.
As shown in FIG. 3A, methylamine, CH3-NH2Including C atoms, N atoms, and H atoms. The C atom and the 3H atoms are respectively connected through C-H chemical bonds, the C atom and the N atom are connected through C-N chemical bonds, and the N atom and the 2H atoms are respectively connected through N-H chemical bonds. The bond lengths of the C-N bond, the C-H bond and the N-H bond are different, and the bond angles formed between the plurality of bonds are different. The atom-chemical bond target sub-map is constructed by utilizing atoms and chemical bonds, and the complete spatial structure information of the molecules to be processed is not enough to be represented.
Fig. 3B schematically shows an atomic-chemical bond target subgraph spectrum of a methylamine molecule according to an embodiment of the disclosure.
As shown in fig. 3B, an Atom-chemical Bond (Atom-Bond) target sub-map 330 may be determined based on the atoms 310 and chemical bonds 320 of a molecule to be treated, such as a methylamine molecule. Atom-chemical bond target sub-graph 330 may be constructed with atoms 310 as nodes and chemical bonds 320 as edges. Atom-chemical bond target sub-map 330 may be used to embody associations and properties between atoms.
Fig. 3C schematically shows a chemical bond-angle target subgraph spectrum of a methylamine molecule according to an embodiment of the disclosure.
As shown in fig. 3C, a chemical Bond-Bond Angle (Bond-Angle) target sub-map 350 may be determined based on the chemical bonds 320 and Bond angles 340 of the molecules to be processed. The chemical bond-bond angle target sub-map 350 may be constructed with the chemical bonds 320 as nodes and the bond angles 340 as edges, for embodying the relationships and properties between the chemical bonds.
Fig. 3D schematically shows a bond angle-dihedral angle target subgraph spectrum schematic of a methylamine molecule according to an embodiment of the disclosure.
As shown in fig. 3D, a bond Angle-Dihedral (Angle-Dihedral) object sub-map 370 can be determined based on the bond Angle 340 and the Dihedral Angle 360 of the molecule to be processed. The key angle-dihedral angle object sub-map 370 may be constructed with the key angles 340 as nodes and the dihedral angles 360 as edges, for embodying the relationship and properties between the key angles.
According to the embodiment of the disclosure, the spatial structure information and the property information of molecules to be processed can be embodied from different dimensions through a plurality of target sub-maps such as an atom-chemical bond target sub-map, a chemical bond-bond angle target sub-map and a bond angle-dihedral angle target sub-map, so that the spatial structure feature representation of the molecules to be processed determined based on the target maps contains abundant molecular spatial structure information and property information, and the prediction of biochemical attribute information related to the spatial structure information based on the spatial structure feature representation is more accurate and faster. For example, the target map includes, for example, a key angle-dihedral angle target sub-map, so that the spatial structure feature representation can represent information of dihedral angles formed between three adjacent chemical bonds inside the molecule to be processed, thereby enabling the spatial structure feature representation to more accurately capture spatial structure information such as isomers and the like, avoiding the problem of limitation in spatial structure feature representation for capturing spatial structure information of the molecule to be processed due to incomplete target map information, and further improving the generalization capability of the application of the spatial structure feature representation.
According to an embodiment of the present disclosure, for operation S220, determining a spatial structural feature representation of the molecule to be processed based on the target profile may include: and inputting the target map into the feature extraction model to obtain the spatial structure feature representation of the molecules to be processed.
According to the embodiment of the disclosure, the feature extraction model may include a Graph network (GNN) or a fully connected network (transform), but is not limited thereto, and may further include a network adapted to a plurality of target sub-maps, Enhanced Geo GNN, constructed based on the Graph network.
According to the embodiment of the disclosure, a plurality of target sub-maps in a target map are respectively used for representing spatial structure information of different dimensions of a molecule, and the feature extraction model may include cascaded M layers of feature extraction sub-models, each feature extraction sub-model includes N feature extraction modules in one-to-one correspondence with N target sub-maps, where N is an integer greater than or equal to 2, and M is an integer greater than or equal to 1. And matching the plurality of target sub-maps by utilizing a plurality of feature extraction modules in the feature extraction sub-model so as to realize the extraction of the spatial structure information with a plurality of different dimensions.
According to an embodiment of the present disclosure, inputting the target profile into the feature extraction model to obtain the spatial structure feature representation of the molecule to be processed may include the following operations.
For example, for an M +1 th layer feature extraction submodel in the feature extraction model, if N is not equal to 1, an M +1 th layer nth feature extraction module is utilized to determine an M +1 th layer nth target sub-feature vector based on a feature vector of an M +1 th layer nth target sub-atlas and an M +1 th layer N +1 th target sub-feature vector, wherein M is 1, and N is 2. And determining the (m + 1) th layer target feature vector based on the (m + 1) th layer 2 nd target sub-feature vector and the m layer target feature vector by using the (m + 1) th layer 1 st feature extraction module. And taking the M-th layer target feature vector as a spatial structure feature representation of the molecule to be processed.
According to embodiments of the present disclosure, the feature extraction model may include cascaded M-layer feature extraction submodels. The first-level feature extraction submodel may be utilized to capture features and spatial structure information of respective first-order neighbors of a plurality of nodes in the target graph, the second-level feature extraction submodel may be utilized to capture features and spatial structure information of respective second-order neighbors of a plurality of nodes in the target graph, by analogy, the characteristics and the spatial structure of respective n-order neighbors of a plurality of nodes in the target map are captured by utilizing the cascaded M-layer characteristic extraction submodels, therefore, high-order three-dimensional space structure characteristic information can be obtained through capturing and modeling, and abundant and complex space structure information is merged into the space structure characteristic representation, so that the space structure characteristic representation can represent more comprehensive information, and then, various prediction tasks such as judging molecular toxicity, identifying targeted drugs, predicting drug combination and the like are performed by utilizing the spatial structure characteristic representation, so that the method is more accurate.
According to embodiments of the present disclosure, the target atlas may include N target sub-atlases. N may be an integer greater than or equal to 2, for example N is an integer of 2, 3, 4, or greater. In the case where N is 2, the target map may include a 1 st target sub-map and a 2 nd target sub-map. In the case where N is 3, the target atlas may include a 1 st target sub-atlas, a 2 nd target sub-atlas, and a 3 rd target sub-atlas. For example, the 1 st target sub-graph comprises an atom-chemical bond target sub-graph. The 2 nd target sub-map comprises a chemical bond-bond angle target sub-map. The 3 rd target sub-map comprises a key angle-dihedral angle target sub-map.
According to the embodiment of the disclosure, the (m + 1) th feature extraction module comprises an (m + 1) th input layer (Embedding), an (m + 1) th graph network layer and an (m + 1) th residual network layer which are cascaded, wherein n is greater than 1. For example, a 2 nd target sub-map, such as a chemical bond-bond angle target sub-map, may be processed using an m +1 th layer n +1 th feature extraction module, and a 3 rd target sub-map, such as a bond angle-dihedral target sub-map, may be processed using an m +1 th layer n +1 th feature extraction module.
According to an embodiment of the present disclosure, determining, by the nth feature extraction module at the (m + 1) th layer, the nth target sub-feature vector at the (m + 1) th layer based on the feature vector of the nth target sub-map at the (m + 1) th layer and the (m + 1) th target sub-feature vector at the (m + 1) th layer includes: and inputting the nth target sub-map of the (m + 1) th layer into the nth input layer of the (m + 1) th layer to obtain the characteristic vector of the nth target sub-map of the (m + 1) th layer. Inputting the feature vector of the nth target sub-map of the (m + 1) th layer and the (n + 1) th target sub-feature vector of the (m + 1) th layer into the nth map network layer of the (m + 1) th layer to obtain the nth initial sub-feature vector of the (m + 1) th layer. And inputting the feature vector of the nth target sub-map of the (m + 1) th layer and the nth initial sub-feature vector of the (m + 1) th layer into the nth residual error network layer of the (m + 1) th layer to obtain the nth target sub-feature vector of the (m + 1) th layer.
According to the embodiment of the disclosure, the nth target sub-map of the (m + 1) th layer is input into the nth input layer of the (m + 1) th layer by using the nth feature extraction module of the (m + 1) th layer, so as to obtain the feature vector of the nth target sub-map of the (m + 1) th layer. And inputting the feature vector of the nth target sub-map of the (m + 1) th layer into the nth map network layer of the (m + 1) th layer to obtain the nth initial sub-feature vector of the (m + 1) th layer. Inputting the feature vector of the nth target sub-map of the (m + 1) th layer and the nth initial sub-feature vector of the (m + 1) th layer into the nth residual error network layer of the (m + 1) th layer to obtain the nth target sub-feature vector of the (m + 1) th layer.
According to the embodiment of the disclosure, the (m + 1) th feature extraction module comprises an (m + 1) th layer 1 figure network layer and an (m + 1) th layer 1 residual error network layer which are cascaded. For example, a 1 st target sub-graph, such as an atom-chemical bond target sub-graph, may be processed with the m +1 st layer 1 feature extraction module.
According to an embodiment of the present disclosure, determining, by the (m + 1) th layer 1 st feature extraction module, the (m + 1) th layer target sub-feature vector and the (m + 1) th layer target feature vector based on the (m + 1) th layer 2 nd target sub-feature vector and the (m + 1) th layer target feature vector includes: and obtaining a 1 st initial sub-feature vector of the m +1 st layer based on the 2 nd target sub-feature vector of the m +1 st layer and the m-th target feature vector by utilizing the 1 st network layer of the m +1 st layer. And obtaining the target feature vector of the (m + 1) th layer based on the (m + 1) th initial sub-feature vector of the (m + 1) th layer and the target feature vector of the (m + 1) th layer by utilizing the (m + 1) th residual error network layer.
According to an embodiment of the present disclosure, the layer 1, feature extraction module may be cascaded to a layer 1, fig. network layer, and layer 1, residual network layer. By using the 1 st layer 1 feature extraction module, the 1 st layer 1 input layer can be used to obtain the feature vector of the 1 st target sub-map based on the 1 st target sub-map. And obtaining a 1 st initial sub-feature vector of the 1 st layer based on the 2 nd target sub-feature vector of the 1 st layer and the feature vector of the 1 st target sub-map by utilizing the 1 st layer 1 and the 1 st network layer. And obtaining a layer 1 target feature vector by utilizing the layer 1 residual error network layer based on the layer 1 initial sub-feature vector and the feature vector of the layer 1 target sub-map.
According to the embodiment of the disclosure, a plurality of target sub-maps with different dimensions can be respectively processed by utilizing a plurality of feature extraction modules in the feature extraction sub-model, and spatial structure information of a plurality of different dimensions is fused to obtain a target sub-feature vector. In addition, a plurality of layers of feature extraction submodels are combined to capture and obtain higher-order three-dimensional space structure feature information, and therefore the space structure feature is expressed and extracted to complete space structure information.
Fig. 4 schematically shows a network structure diagram of a feature extraction model according to an embodiment of the present disclosure.
As shown in fig. 4, the feature extraction model includes 2-level feature extraction submodels, such as a level 1 feature extraction submodel 410 and a level 2 feature extraction submodel 420. Each layer of feature extraction submodel comprises 3 feature extraction modules, such as a layer 1, a layer 2 and a layer 1, a layer 3.
For the layer 1 feature extraction submodel 410 in the feature extraction model, the key angle-dihedral angle target sub-map 450 is input into the layer 1 input layer 14131 in the layer 1 layer 3 feature extraction module, and the feature vector of the layer 1 layer 3 target sub-map is obtained. And inputting the feature vector of the 3 rd target sub-map of the 1 st layer into the 3 rd network layer GNN 14132 of the 1 st layer to obtain a 3 rd initial sub-feature vector of the 1 st layer. Inputting the feature vector of the 3 rd target sub-map of the 1 st layer and the 3 rd initial sub-feature vector of the 1 st layer into the 3 rd residual network layer 4133 of the 1 st layer to obtain a 3 rd target sub-feature vector of the 1 st layer.
The chemical bond-bond angle target sub-map 440 is input into the layer 1 and layer 2 input layer 14121 in the layer 1 and layer 2 feature extraction module to obtain the feature vector of the layer 1 and layer 2 target sub-map. Inputting the feature vector of the 2 nd target sub-map of the 1 st layer and the 3 rd target sub-feature vector of the 1 st layer into the 2 nd network layer GNN 14122 of the 1 st layer to obtain the 2 nd initial sub-feature vector of the 1 st layer. Inputting the feature vector of the layer 1, layer 2 target sub-map and the layer 1, layer 2 initial sub-feature vector into the layer 1, layer 2 residual error network layer 4123 to obtain a layer 1, layer 2 target sub-feature vector.
Inputting the atom-chemical bond target sub-map 430 into a layer 1 input layer 14111 in a layer 1 feature extraction module to obtain a feature vector of the layer 1 target sub-map. Inputting the feature vector of the 1 st target sub-map of the 1 st layer and the 2 nd target sub-feature vector of the 1 st layer into the 1 st network layer GNN 14112 of the 1 st layer and the 1 st figure to obtain the 1 st initial sub-feature vector of the 1 st layer. Inputting the feature vector of the 1 st target sub-map of the 1 st layer and the 1 st initial sub-feature vector of the 1 st layer into the 1 st residual error network layer 4113 of the 1 st layer to obtain the 1 st target feature vector of the 1 st layer.
For the layer 2 feature extraction submodel in the feature extraction model, the key angle-dihedral angle target sub-map 450 is input into the layer 2 input layer 24231 in the layer 2 and layer 3 feature extraction module, and the feature vector of the layer 2 and layer 3 target sub-map is obtained. And inputting the feature vector of the 3 rd target sub-map of the 2 nd layer into the 3 rd network layer GNN 24232 of the 2 nd layer to obtain a 3 rd initial sub-feature vector of the 2 nd layer. Inputting the feature vector of the 3 rd target sub-map of the 2 nd layer and the 3 rd initial sub-feature vector of the 2 nd layer into the 3 rd residual network layer 4233 of the 2 nd layer to obtain a 3 rd target sub-feature vector of the 2 nd layer.
The chemical bond-bond angle target sub-map 440 is input into the layer 2 input layer 24221 in the layer 2 feature extraction module to obtain the feature vector of the layer 2 target sub-map. And inputting the feature vector of the 2 nd target sub-map of the 2 nd layer and the 3 rd target sub-feature vector of the 2 nd layer into the 2 nd layer and the 2 nd map network layer GNN 24222 to obtain a 2 nd layer and 2 nd initial sub-feature vector. Inputting the feature vector of the layer 2, layer 2 target sub-map and the layer 2 initial sub-feature vector into the layer 2, layer 2 residual error network layer 4223, and obtaining a layer 2, layer 2 target sub-feature vector.
Inputting the layer 1, the target feature vector and the layer 2, the target sub-feature vector into the layer 2, the layer 1, the network layer GNN 24212 to obtain a layer 2, the layer 1, the initial sub-feature vector. Inputting the layer 2 layer 1 initial sub-feature vector and the layer 2 layer 1 initial sub-feature vector into the layer 2 layer 1 residual error network layer 4213 to obtain a layer 2 layer 1 target feature vector. The layer 2, layer 1, target feature vectors are taken as the spatial structure feature representation 460.
It should be noted that the embodiment shown in fig. 4 is only an exemplary embodiment, and m may be 1, but is not limited thereto, and may also be any integer such as 2, 3, and the like. However, the more the number of layers of the feature extraction submodel is, the larger the data processing amount is, the more complete the spatial structure information captured by the spatial structure feature representation is, and the number of m can be determined according to actual needs. In addition, n may be 2 or 3, but is not limited thereto, and may be other integer greater than 1. Can be determined according to the number of target sub-maps, and is not described in detail herein.
FIG. 5 schematically shows a 2-butene molecular structure schematic according to an embodiment of the present disclosure.
As shown in FIG. 5, 2-butene includes two stereo structures, such as cis-2-butene and trans-2-butene, which are isomers of each other.
According to the embodiment of the disclosure, the spatial structure feature representation of the molecule to be processed is determined by using the target map comprising the atom-chemical bond target sub-map, the chemical bond-bond angle target sub-map and the bond angle-dihedral angle target sub-map, so that the spatial structure feature representation captures a more complete three-dimensional spatial structure of the molecule to be processed, for example, dihedral angle information formed between three adjacent chemical bonds of the molecule to be processed is used, thereby enabling the spatial structure feature representation to represent fine and important spatial structure information such as information of isomers, and further improving the generalization performance of the spatial structure feature representation.
Fig. 6 schematically shows a flow chart of a training method of a feature extraction model according to an embodiment of the present disclosure.
As shown in fig. 6, the method includes operations S610 to S640.
In operation S610, a target sample pattern is determined using molecular structure information of sample molecules, wherein the target sample pattern is used to characterize a plurality of molecular spatial structure information of different dimensions.
In operation S620, the target sample atlas is processed using the feature extraction model to determine a sample spatial structure feature representation of the sample molecules.
In operation S630, attribute prediction information of the sample molecules is determined based on the sample spatial structure feature representation.
In operation S640, the feature extraction model is trained by using the attribute prediction information of the sample molecules and the labels of the sample molecules, so as to obtain a trained feature extraction model, where the labels are used to represent the attribute information of the sample molecules.
According to an embodiment of the present disclosure, a sample spatial structure feature representation may be processed using a pre-trained molecular property prediction model to obtain property prediction information of sample molecules.
According to the embodiment of the present disclosure, the attribute prediction information and the label may not be limited, and for example, the attribute information may be solubility, toxicity, absorption, metabolism, and the like of the sample molecule, but the attribute prediction information and the label may be determined according to actual conditions as long as they are the same.
According to the embodiment of the disclosure, the feature extraction model is trained by using the target sample maps for representing the plurality of molecular spatial structure information with different dimensions, so that the trained feature extraction model can be suitable for extracting features from the target maps with the plurality of molecular spatial structure information with different dimensions, and spatial structure feature representation with complete information is obtained.
According to an embodiment of the present disclosure, a plurality of target subsample patterns of the target sample pattern are each used for characterizing spatial structure information of different dimensions of the molecule, and the feature extraction model may include cascaded M-layer feature extraction submodels, each feature extraction submodel including N feature extraction modules in one-to-one correspondence with N target subsample patterns, where N is an integer greater than or equal to 2, and M is an integer greater than or equal to 1.
According to an embodiment of the present disclosure, inputting the target sample profile into the feature extraction model to obtain the spatial structure feature representation of the molecule to be processed may include the following operations.
For example, for an M +1 th layer feature extraction submodel in the feature extraction model, in the case that N is not equal to 1, an M +1 th layer nth feature extraction module is utilized to determine an M +1 th layer nth target sub-feature vector based on a feature vector of an M +1 th layer nth target sub-sample atlas and an M +1 th layer N +1 th target sub-feature vector, wherein M is 1, a. And determining the (m + 1) th layer target feature vector based on the (m + 1) th layer 2 nd target sub-feature vector and the m layer target feature vector by using the (m + 1) th layer 1 st feature extraction module. And taking the M-th layer target feature vector as a spatial structure feature representation of the molecule to be processed.
According to an embodiment of the present disclosure, the target sample atlas may include N target sub-sample atlases. N may be an integer greater than or equal to 2, for example N is an integer of 2, 3, 4, or greater. In the case where N is 2, the target sample pattern may include a 1 st target sub-sample pattern and a 2 nd target sub-sample pattern. In the case where N is 3, the target sample pattern may include a 1 st target sub-sample pattern, a 2 nd target sub-sample pattern, and a 3 rd target sub-sample pattern. For example, the 1 st target subsample map comprises an atom-chemical bond target subsample map. The 2 nd target subsample pattern comprises a chemical bond-bond angle target subsample pattern. The 3 rd target subsample map comprises a key angle-dihedral target subsample map.
According to the embodiment of the disclosure, the (m + 1) th layer n feature extraction module comprises an (m + 1) th input layer, an (m + 1) th graph network layer and an (m + 1) th residual error network layer which are cascaded, wherein n is greater than 1. For example, the 2 nd target sub-sample pattern, e.g., the chemical bond-bond angle target sub-sample pattern, may be processed using the m +1 th layer n +1 th feature extraction module, and the 3 rd target sub-sample pattern, e.g., the bond angle-dihedral angle target sub-sample pattern, may be processed using the m +1 th layer n +1 th feature extraction module.
According to an embodiment of the present disclosure, determining, by the nth feature extraction module at the (m + 1) th layer, the nth target sub-sample feature vector at the (m + 1) th layer based on the feature vector of the nth target sub-sample atlas at the (m + 1) th layer and the (m + 1) th target sub-sample feature vector at the (m + 1) th layer includes: and inputting the nth target sub-sample atlas of the (m + 1) th layer into the nth input layer of the (m + 1) th layer to obtain the feature vector of the nth target sub-sample atlas of the (m + 1) th layer. Inputting the feature vector of the nth target sub-sample map of the (m + 1) th layer and the (n + 1) th target sub-sample feature vector of the (m + 1) th layer into the nth map network layer of the (m + 1) th layer to obtain the nth initial sub-sample feature vector of the (m + 1) th layer. And inputting the feature vector of the nth target subsample atlas at the (m + 1) th layer and the nth initial subsample feature vector at the (m + 1) th layer into the nth residual error network layer at the (m + 1) th layer to obtain the nth target subsample feature vector at the (m + 1) th layer.
According to the embodiment of the disclosure, the (m + 1) th feature extraction module comprises an (m + 1) th layer 1 figure network layer and an (m + 1) th layer 1 residual error network layer which are cascaded. For example, a 1 st target subsample map, e.g., an atom-chemical bond target subsample map, may be processed with an m +1 st layer 1 feature extraction module.
According to an embodiment of the present disclosure, determining, by the (m + 1) th layer 1 st feature extraction module, the (m + 1) th layer 2 nd target sub-sample feature vector and the (m) th layer target sample feature vector includes: and obtaining a 1 st initial sub-sample feature vector of the m +1 st layer based on the 2 nd target sub-sample feature vector of the m +1 st layer and the target sample feature vector of the m < th > layer by utilizing the 1 st graph network layer of the m +1 st layer. And obtaining the (m + 1) th layer target sample feature vector based on the (m + 1) th initial sub-sample feature vector and the (m) th layer target sample feature vector by using the (m + 1) th layer 1 st residual error network layer.
Fig. 7 schematically illustrates a flow diagram of molecular property identification according to an embodiment of the present disclosure.
As shown in fig. 7, the method includes operations S710 to S720.
In operation S710, the molecules to be recognized are processed using a data processing method, and a spatial structural feature representation of the molecules to be recognized is determined.
In operation S720, target attribute information of the molecule to be recognized is determined based on the spatial structure feature representation.
According to an embodiment of the present disclosure, the target property information of the molecule to be recognized may include absorption, distribution, metabolism, excretion, toxicity, but is not limited thereto, and the target property information of the molecule to be recognized may also be utilized to include property information for Drug combination (DDI), and property information for Diffusion Tensor Imaging (DTI).
According to the embodiment of the disclosure, the target attribute information of the molecules to be identified, such as drug molecules, is predicted by using the spatial structure characteristic representation bearing multi-dimensional and fine spatial structure information, so that the method can be more accurate and effective, the drug research and development efficiency is improved, the cost is reduced, and the research and development range is expanded.
Fig. 8 schematically shows a block diagram of a data processing apparatus according to an embodiment of the present disclosure.
As shown in fig. 8, the data processing apparatus 800 may include a first determination module 810 and a second determination module 820.
A first determining module 810, configured to determine a target map using the molecular structure information of the molecule to be processed, where the target map is used to represent a plurality of molecular spatial structure information with different dimensions.
A second determination module 820 for determining a spatial structure characteristic representation of the molecule to be processed based on the target profile.
According to an embodiment of the present disclosure, the second determination module may include an extraction sub-module.
And the extraction submodule is used for inputting the target map into the feature extraction model to obtain the spatial structure feature representation of the molecules to be processed.
According to an embodiment of the present disclosure, the target atlas includes N target sub-atlases; the feature extraction model comprises cascaded M layers of feature extraction submodels, each feature extraction submodel comprises N feature extraction modules which are in one-to-one correspondence with N target sub-maps, wherein N is an integer greater than or equal to 2, and M is an integer greater than or equal to 1.
According to an embodiment of the present disclosure, the extraction sub-module may include a first extraction unit, a second extraction unit, and a first determination unit.
A first extraction unit, configured to extract a sub-model for an M +1 th layer of features in the feature extraction model, and if N is not equal to 1, determine an M +1 th layer N target sub-feature vector based on a feature vector of an M +1 th layer N target sub-atlas and an M +1 th layer N +1 th target sub-feature vector by using an M +1 th layer N feature extraction module, where M ═ 1,... An.m-1, N · 2,. An.
And the second extraction unit is used for determining the (m + 1) th layer target feature vector based on the (m + 1) th layer 2 < th > target sub-feature vector and the m < th > layer target feature vector by using the (m + 1) th layer 1 < th > feature extraction module.
And the first determining unit is used for representing the M-th layer target characteristic vector as the spatial structure characteristic of the molecule to be processed.
According to the embodiment of the disclosure, the (m + 1) th feature extraction module comprises an (m + 1) th input layer, an (m + 1) th graph network layer and an (m + 1) th residual error network layer which are cascaded, wherein n is greater than 1.
According to an embodiment of the present disclosure, the first extraction unit may include a first input subunit, a first extraction subunit, and a first fusion subunit.
The first input subunit is used for inputting the (m + 1) th target sub-map into the (m + 1) th input layer to obtain the feature vector of the (m + 1) th target sub-map.
The first extraction subunit is used for inputting the feature vector of the nth target sub-map of the (m + 1) th layer and the (n + 1) th target sub-feature vector of the (m + 1) th layer into the nth map network layer of the (m + 1) th layer to obtain the nth initial sub-feature vector of the (m + 1) th layer.
The first fusion subunit is configured to input the feature vector of the nth target sub-map of the (m + 1) th layer and the nth initial sub-feature vector of the (m + 1) th layer into the nth residual network layer of the (m + 1) th layer, so as to obtain the nth target sub-feature vector of the (m + 1) th layer.
According to the embodiment of the disclosure, the (m + 1) th feature extraction module comprises an (m + 1) th layer 1 figure network layer and an (m + 1) th layer 1 residual error network layer which are cascaded.
According to an embodiment of the present disclosure, the second extraction unit may include a second extraction sub-unit and a second fusion sub-unit.
And the second extraction subunit is used for obtaining the (m + 1) st initial sub-feature vector of the (m + 1) th layer based on the (m + 1) th target sub-feature vector and the (m) th layer target feature vector by using the (m + 1) th layer of figure 1 network layer.
And the second fusion subunit is used for obtaining the (m + 1) th layer target feature vector based on the (m + 1) th initial sub-feature vector and the (m) th layer target feature vector by using the (m + 1) th layer 1 st residual network layer.
According to an embodiment of the present disclosure, the target maps include an atom-chemical bond target sub-map, a chemical bond-bond angle target sub-map, and a bond angle-dihedral angle target sub-map.
According to an embodiment of the present disclosure, the first determination module may include a first map determination unit, a second map determination unit, and a third map determination unit.
A first map determining unit for determining an atom-chemical bond target sub-map based on atoms of the molecule to be processed and chemical bonds of the molecule to be processed.
A second map determining unit for determining a chemical bond-bond angle target sub-map based on the chemical bond of the molecule to be processed and the bond angle of the molecule to be processed.
A third map determining unit for determining a bond angle-dihedral angle target sub-map based on the bond angle of the molecule to be processed and the dihedral angle of the molecule to be processed.
According to an embodiment of the present disclosure, the data processing apparatus may further include, before the first determining module: an expression determination module and a structure determination module.
And the expression determination module is used for determining the molecular expression of the molecule to be processed.
And the structure determining module is used for determining the molecular structure information of the molecules to be processed based on the molecular expression.
Fig. 9 schematically shows a block diagram of a training apparatus of a feature extraction model according to an embodiment of the present disclosure.
As shown in fig. 9, the training apparatus 900 for feature extraction model may include a sample map determination module 910, a sample extraction module 920, a prediction module 930, and a training module 940.
A sample map determining module 910, configured to determine a target sample map by using the molecular structure information of the sample molecules, where the target sample map is used to represent multiple molecular spatial structure information with different dimensions.
And a sample extraction module 920, configured to process the target sample map by using the feature extraction model, and determine a sample spatial structure feature representation of the sample molecule.
A prediction module 930 configured to determine property prediction information of the sample molecules based on the spatial structure feature representation of the sample.
And a training module 940, configured to train the feature extraction model by using the attribute prediction information of the sample molecules and the labels of the sample molecules, to obtain the trained feature extraction model, where the labels are used to represent the attribute information of the sample molecules.
Fig. 10 schematically shows a block diagram of a molecular property identification apparatus according to an embodiment of the present disclosure.
As shown in FIG. 10, the molecular property recognition device 1000 may include a feature representation determination module 1010 and a recognition module 1020.
A feature representation determining module 1010, configured to process the molecules to be identified by using the data processing apparatus of the present disclosure, and determine a spatial structure feature representation of the molecules to be identified.
An identification module 1020 for determining target property information of the molecule to be identified based on the spatial structure feature representation.
The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.
According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method according to an embodiment of the disclosure.
According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium having stored thereon computer instructions for causing a computer to perform a method as in an embodiment of the present disclosure.
According to an embodiment of the disclosure, a computer program product comprising a computer program which, when executed by a processor, implements a method as in an embodiment of the disclosure.
FIG. 11 shows a schematic block diagram of an example electronic device 1100 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 11, the device 1100 comprises a computing unit 1101, which may perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM)1102 or a computer program loaded from a storage unit 1108 into a Random Access Memory (RAM) 1103. In the RAM 1103, various programs and data necessary for the operation of the device 1100 may also be stored. The calculation unit 1101, the ROM 1102, and the RAM 1103 are connected to each other by a bus 1104. An input/output (I/O) interface 1105 is also connected to bus 1104.
A number of components in device 1100 connect to I/O interface 1105, including: an input unit 1106 such as a keyboard, a mouse, and the like; an output unit 1107 such as various types of displays, speakers, and the like; a storage unit 1108 such as a magnetic disk, optical disk, or the like; and a communication unit 1109 such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 1109 allows the device 1100 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.
The computing unit 1101 can be a variety of general purpose and/or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and the like. The calculation unit 1101 performs the respective methods and processes described above, such as a data processing method, a training method of a feature extraction model, or a molecular property recognition method. For example, in some embodiments, the data processing method, the training method of the feature extraction model, or the molecular property recognition method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1108. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 1100 via ROM 1102 and/or communication unit 1109. When the computer program is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the data processing method, the training method of the feature extraction model, or the molecular property recognition method described above may be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured by any other suitable means (e.g., by means of firmware) to perform a data processing method, a training method of a feature extraction model, or a molecular property recognition method.
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), Complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a 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 may be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server with a combined blockchain.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and the present disclosure is not limited herein.
The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.

Claims (21)

1. A method of data processing, comprising:
determining a target map by utilizing molecular structure information of molecules to be processed, wherein the target map is used for representing a plurality of molecular space structure information with different dimensions; and
and determining the spatial structure characteristic representation of the molecules to be processed based on the target map.
2. The method of claim 1, wherein the determining a spatial structural feature representation of the molecule to be processed based on the target profile comprises:
inputting the target map into a feature extraction model to obtain a spatial structure feature representation of the molecules to be processed;
wherein the target atlas comprises N target sub-atlases; the feature extraction model comprises cascaded M layers of feature extraction submodels, each feature extraction submodel comprises N feature extraction modules which are in one-to-one correspondence with the N target sub-maps, wherein N is an integer greater than or equal to 2, and M is an integer greater than or equal to 1.
3. The method of claim 2, wherein the inputting the target profile into a feature extraction model, and the obtaining of the spatial structural feature representation of the molecule to be processed comprises:
for an (M + 1) th layer feature extraction submodel in the feature extraction model, if N is not equal to 1, determining an (M + 1) th layer N target sub-feature vector based on a feature vector of an (M + 1) th layer N target sub-atlas and an (M + 1) th layer N +1 th target sub-feature vector by using an (M + 1) th layer N feature extraction module, wherein M is 1, and N is 2;
determining an m + 1-layer target feature vector based on the m + 1-layer 2-layer target sub-feature vector and the m-layer target feature vector by using an m + 1-layer 1-th feature extraction module; and
and taking the M-th layer target feature vector as the spatial structure feature representation of the molecule to be processed.
4. The method according to claim 3, wherein the (m + 1) th layer nth feature extraction module comprises (m + 1) th input layer, an (m + 1) th graph network layer, and an (m + 1) th residual network layer which are cascaded, wherein n is greater than 1;
the determining, by using the (m + 1) th feature extraction module, the (m + 1) th target sub-feature vector based on the feature vector of the (m + 1) th target sub-map and the (m + 1) th target sub-feature vector of the (m + 1) th layer includes:
inputting the nth target sub-map of the (m + 1) th layer into the nth input layer of the (m + 1) th layer to obtain the characteristic vector of the nth target sub-map of the (m + 1) th layer;
inputting the feature vector of the nth target sub-map of the (m + 1) th layer and the (n + 1) th target sub-feature vector of the (m + 1) th layer into the nth map network layer of the (m + 1) th layer to obtain an nth initial sub-feature vector of the (m + 1) th layer; and
inputting the feature vector of the nth target sub-map of the (m + 1) th layer and the nth initial sub-feature vector of the (m + 1) th layer into the nth residual error network layer of the (m + 1) th layer to obtain the nth target sub-feature vector of the (m + 1) th layer.
5. The method according to claim 3 or 4, wherein the (m + 1) th feature extraction module comprises (m + 1) th layer fig. 1 network layer, (m + 1) th layer 1 residual network layer in cascade;
the determining, by the (m + 1) th layer 1 st feature extraction module, the (m + 1) th layer target feature vector based on the (m + 1) th layer 2 nd target sub-feature vector and the m layer target feature vector includes:
obtaining a 1 st initial sub-feature vector of an m +1 th layer based on the 2 nd target sub-feature vector of the m +1 th layer and the m-th layer target feature vector by using a 1 st network layer of the m +1 th layer; and
and obtaining the (m + 1) th layer target feature vector by using the (m + 1) th layer 1 st residual error network layer based on the (m + 1) th layer 1 st initial sub-feature vector and the m-th layer target feature vector.
6. The method of any one of claims 2 to 5, wherein the target patterns include an atom-chemical bond target sub-pattern, a chemical bond-bond angle target sub-pattern, and a bond angle-dihedral angle target sub-pattern;
the determining the target map by using the molecular structure information of the molecules to be processed comprises the following steps:
determining the atom-chemical bond target sub-map based on the atoms of the molecules to be treated and the chemical bonds of the molecules to be treated;
determining the chemical bond-bond angle target sub-map based on the chemical bond of the molecule to be treated and the bond angle of the molecule to be treated; and
determining the bond angle-dihedral angle target sub-map based on the bond angle of the molecule to be treated and the dihedral angle of the molecule to be treated.
7. The method of any one of claims 1 to 6, further comprising, prior to said determining a target profile using molecular structure information of a molecule to be processed:
determining a molecular expression of the molecule to be processed; and
and determining the molecular structure information of the molecules to be processed based on the molecular expression.
8. A training method of a feature extraction model comprises the following steps:
determining a target sample map by using the molecular structure information of the sample molecules, wherein the target sample map is used for representing a plurality of molecular space structure information with different dimensions;
processing the target sample map by using a feature extraction model, and determining sample space structure feature representation of the sample molecules;
determining attribute prediction information of the sample molecules based on the sample spatial structure feature representation; and
and training the feature extraction model by using the attribute prediction information of the sample molecules and the labels of the sample molecules to obtain the trained feature extraction model, wherein the labels are used for representing the attribute information of the sample molecules.
9. A molecular attribute identification method comprises the following steps:
processing a molecule to be identified by using the method according to any one of claims 1 to 7, determining a spatial structural feature representation of the molecule to be identified; and
and determining target attribute information of the molecules to be identified based on the spatial structure characteristic representation.
10. A data processing apparatus comprising:
the device comprises a first determination module, a second determination module and a third determination module, wherein the first determination module is used for determining a target map by utilizing molecular structure information of molecules to be processed, and the target map is used for representing a plurality of molecular space structure information with different dimensions; and
a second determination module for determining a spatial structure characteristic representation of the molecule to be processed based on the target profile.
11. The apparatus of claim 10, wherein the second determining means comprises:
the extraction submodule is used for inputting the target map into a feature extraction model to obtain the spatial structure feature representation of the molecules to be processed;
wherein the target atlas comprises N target sub-atlases; the feature extraction model comprises cascaded M layers of feature extraction submodels, each feature extraction submodel comprises N feature extraction modules which are in one-to-one correspondence with the N target sub-maps, wherein N is an integer greater than or equal to 2, and M is an integer greater than or equal to 1.
12. The apparatus of claim 11, wherein the extraction submodule comprises:
a first extraction unit, configured to, for an (M + 1) th layer feature extraction submodel in the feature extraction model, determine, by an (M + 1) th layer N-th feature extraction module, an (M + 1) th layer N-th target sub-feature vector based on a feature vector of an (M + 1) th layer N-th target sub-atlas and an (M + 1) th layer N + 1-th target sub-feature vector if N is not equal to 1, where M is 1, N, M-1, N is 2, N;
the second extraction unit is used for determining an m + 1-layer target feature vector based on the m + 1-layer 2-layer target sub-feature vector and the m-layer target feature vector by using the m + 1-layer feature extraction module; and
and the first determining unit is used for representing the M-th layer target characteristic vector as the spatial structure characteristic of the molecule to be processed.
13. The apparatus according to claim 12, wherein the (m + 1) th feature extraction module comprises (m + 1) th input layer, an (m + 1) th graph network layer, and an (m + 1) th residual network layer, which are cascaded, wherein n is greater than 1;
the first extraction unit includes:
the first input subunit is used for inputting the (m + 1) th target sub-map into the (m + 1) th input layer to obtain the feature vector of the (m + 1) th target sub-map;
the first extraction subunit is configured to input the feature vector of the nth target sub-map of the (m + 1) th layer and the nth +1 target sub-feature vector of the (m + 1) th layer into the nth map network layer of the (m + 1) th layer, so as to obtain an nth initial sub-feature vector of the (m + 1) th layer; and
the first fusion sub-unit is configured to input the feature vector of the nth target sub-map of the (m + 1) th layer and the nth initial sub-feature vector of the (m + 1) th layer into the nth residual network layer of the (m + 1) th layer, so as to obtain the nth target sub-feature vector of the (m + 1) th layer.
14. The apparatus according to claim 12 or 13, wherein the (m + 1) th feature extraction module comprises (m + 1) th layer fig. 1 network layer, and (m + 1) th layer 1 residual network layer in cascade;
the second extraction unit includes:
the second extraction subunit is configured to, by using the (m + 1) th layer of fig. 1 network layer, obtain a (m + 1) th initial sub-feature vector based on the (m + 1) th layer of 2 nd target sub-feature vector and the (m) th layer of target feature vector; and
and the second fusion subunit is configured to obtain the (m + 1) th layer target feature vector based on the (m + 1) th initial sub-feature vector and the (m + 1) th layer target feature vector by using the (m + 1) th layer 1 st residual network layer.
15. The apparatus of any one of claims 11 to 14, wherein the target patterns comprise an atom-chemical bond target sub-pattern, a chemical bond-bond angle target sub-pattern, and a bond angle-dihedral angle target sub-pattern;
the first determining module includes:
a first map determining unit for determining the atom-chemical bond target sub-map based on the atoms of the molecules to be processed and the chemical bonds of the molecules to be processed;
a second map determining unit for determining the chemical bond-bond angle target sub-map based on the chemical bond of the molecule to be processed and the bond angle of the molecule to be processed; and
a third map determining unit for determining the bond angle-dihedral angle target sub-map based on the bond angle of the molecule to be processed and the dihedral angle of the molecule to be processed.
16. The apparatus of any of claims 10-15, further comprising, prior to the first determining means:
the expression determining module is used for determining the molecular expression of the molecule to be processed; and
and the structure determining module is used for determining the molecular structure information of the molecules to be processed based on the molecular expression.
17. A training apparatus for a feature extraction model, comprising:
the sample map determining module is used for determining a target sample map by utilizing the molecular structure information of sample molecules, wherein the target sample map is used for representing a plurality of molecular space structure information with different dimensions;
the sample extraction module is used for processing the target sample map by using a feature extraction model and determining the sample space structure feature representation of the sample molecules;
a prediction module for determining attribute prediction information of the sample molecules based on the sample spatial structure feature representation; and
and the training module is used for training the feature extraction model by utilizing the attribute prediction information of the sample molecules and the labels of the sample molecules to obtain the trained feature extraction model, wherein the labels are used for representing the attribute information of the sample molecules.
18. A molecular property recognition apparatus, comprising:
a feature representation determination module for processing a molecule to be identified using the data processing apparatus according to any of claims 10 to 16, determining a spatial structural feature representation of the molecule to be identified; and
and the identification module is used for determining the target attribute information of the molecules to be identified based on the spatial structure feature representation.
19. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the data processing method of any one of claims 1 to 7, the training method of the feature extraction model of claim 8, or the molecular property recognition method of claim 9.
20. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the data processing method according to any one of claims 1 to 7, the training method of the feature extraction model according to claim 8, or the molecular property recognition method according to claim 9.
21. A computer program product comprising a computer program which, when executed by a processor, implements the data processing method of any one of claims 1 to 7, the training method of the feature extraction model of claim 8, or the molecular property recognition method of claim 9.
CN202210097241.7A 2022-01-26 2022-01-26 Data processing method, training method, recognition method, device, equipment and medium Pending CN114429801A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210097241.7A CN114429801A (en) 2022-01-26 2022-01-26 Data processing method, training method, recognition method, device, equipment and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210097241.7A CN114429801A (en) 2022-01-26 2022-01-26 Data processing method, training method, recognition method, device, equipment and medium

Publications (1)

Publication Number Publication Date
CN114429801A true CN114429801A (en) 2022-05-03

Family

ID=81313976

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210097241.7A Pending CN114429801A (en) 2022-01-26 2022-01-26 Data processing method, training method, recognition method, device, equipment and medium

Country Status (1)

Country Link
CN (1) CN114429801A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115662534A (en) * 2022-12-14 2023-01-31 药融云数字科技(成都)有限公司 Chemical structure determination method and system based on map, storage medium and terminal

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115662534A (en) * 2022-12-14 2023-01-31 药融云数字科技(成都)有限公司 Chemical structure determination method and system based on map, storage medium and terminal

Similar Documents

Publication Publication Date Title
CN109074607B (en) Learned model providing method and learned model providing device
CN114357105B (en) Pre-training method and model fine-tuning method of geographic pre-training model
CN113947147B (en) Training method, positioning method and related device of target map model
CN112784778A (en) Method, apparatus, device and medium for generating model and identifying age and gender
CN111881777B (en) Video processing method and device
CN113642583B (en) Deep learning model training method for text detection and text detection method
CN113360711B (en) Model training and executing method, device, equipment and medium for video understanding task
WO2020148588A1 (en) Machine learning framework for finding materials with desired properties
CN113705628B (en) Determination method and device of pre-training model, electronic equipment and storage medium
CN115455161A (en) Conversation processing method, conversation processing device, electronic equipment and storage medium
CN110633717A (en) Training method and device for target detection model
CN114817612A (en) Method and related device for calculating multi-modal data matching degree and training calculation model
CN112784102B (en) Video retrieval method and device and electronic equipment
CN114429801A (en) Data processing method, training method, recognition method, device, equipment and medium
CN117670366A (en) Risk prediction method, apparatus, device, medium, and program product
CN113255770B (en) Training method of compound attribute prediction model and compound attribute prediction method
CN116796282A (en) Molecular screening method, training device, electronic equipment and storage medium
CN113255769B (en) Training method of compound attribute prediction model and compound attribute prediction method
CN113515687B (en) Logistics information acquisition method and device
CN114332509A (en) Image processing method, model training method, electronic device and automatic driving vehicle
CN112765975B (en) Word segmentation disambiguation processing method, device, equipment and medium
CN116563665B (en) Training method of target detection model, target detection method, device and equipment
WO2023236900A1 (en) Item recommendation method and related device thereof
CN114491319A (en) Information sorting method and device, electronic equipment and storage medium
CN115761430A (en) Target detection method, model training method, device, medium and electronic equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination