CN117393075A - Model training method and task execution method based on molecular energy information - Google Patents
Model training method and task execution method based on molecular energy information Download PDFInfo
- Publication number
- CN117393075A CN117393075A CN202311702625.8A CN202311702625A CN117393075A CN 117393075 A CN117393075 A CN 117393075A CN 202311702625 A CN202311702625 A CN 202311702625A CN 117393075 A CN117393075 A CN 117393075A
- Authority
- CN
- China
- Prior art keywords
- molecular
- information
- molecule
- specified compound
- dimensional
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000012549 training Methods 0.000 title claims abstract description 53
- 150000001875 compounds Chemical class 0.000 claims abstract description 232
- 238000010586 diagram Methods 0.000 claims abstract description 74
- 238000012512 characterization method Methods 0.000 claims abstract description 50
- 230000007246 mechanism Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 16
- 238000003860 storage Methods 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 14
- 238000013461 design Methods 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 21
- 230000006870 function Effects 0.000 description 13
- 238000011160 research Methods 0.000 description 6
- 238000011161 development Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004925 denaturation Methods 0.000 description 2
- 230000036425 denaturation Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/27—Regression, e.g. linear or logistic regression
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Crystallography & Structural Chemistry (AREA)
- Databases & Information Systems (AREA)
- Chemical & Material Sciences (AREA)
- Medical Informatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The specification discloses a model training method and a task execution method based on molecular energy information. The model training method comprises the following steps: acquiring characterization data of a specified compound molecule, and determining three-dimensional molecular map information capable of specifying the compound molecule; inputting the three-dimensional molecular diagram information into a prediction model, and determining the three-dimensional molecular diagram characteristics corresponding to the specified compound molecules; predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular diagram characteristics; the model is trained with the objective of optimizing the minimized deviation between the predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule. In the subsequent process of predicting the energy of the whole molecule by the molecular structure characteristics, the whole energy of the compound molecule under any structure can be rapidly and accurately predicted by the prediction model, and the accuracy of molecular energy prediction is improved.
Description
Technical Field
The specification relates to the technical field of computers, in particular to a model training method and a task execution method based on molecular energy information.
Background
With the development of science and technology, molecular research has become an important part of the social and scientific development process, and by constructing the relationship between the molecular structure and energy of a compound, the molecular research can provide a favorable technical guarantee for the research of downstream tasks (molecular regulation, molecular design and the like) related to the configuration change of the compound molecule, so that the efficient and accurate prediction of the energy of the compound molecule becomes a new exploration direction in the molecular research process.
However, at present, a simple shallow neural network is generally used for describing compound molecules, the accuracy of molecular energy of the compound predicted by a model is low, and the compound is difficult to be used for performing downstream molecular research tasks.
Therefore, how to accurately determine the energy information of the compound molecule is a urgent problem to be solved.
Disclosure of Invention
The present specification provides a model training method and a task execution method based on molecular energy information, so as to partially solve the above-mentioned problems existing in the prior art.
The technical scheme adopted in the specification is as follows:
the present specification provides a method of model training, comprising:
acquiring characterization data of a specified compound molecule, wherein the characterization data are used for characterizing position information and attribute information of each atom in the specified compound molecule;
Processing the characterization data to determine three-dimensional molecular map information of the specified compound molecules;
inputting the three-dimensional molecular diagram information into a prediction model to be trained, and determining three-dimensional molecular diagram characteristics corresponding to the specified compound molecules according to the three-dimensional molecular diagram information based on isomorphism between the position information and the embedded characteristics corresponding to the position information and invariance between the attribute information and the embedded characteristics corresponding to the attribute information through the prediction model;
predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular graph characteristics;
and training the prediction model by taking the deviation between the predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule as an optimization target.
Optionally, obtaining characterization data of the specified compound molecule specifically includes:
selecting initial data for the specified compound molecule in a dataset of molecular compounds;
determining the characterization data based on the initial data for the specified compound molecule.
Optionally, the location information includes: coordinates of each atom in the molecule of the specified compound under a specified coordinate system;
The attribute information includes: the type of each atom in the molecule of the specified compound, the direction vector between any two atoms in the molecule of the specified compound and the connection information between any two atoms in the molecule of the specified compound.
Optionally, a graph annotation force mechanism network is arranged in the prediction model;
determining the three-dimensional molecular diagram characteristics corresponding to the specified compound molecules according to the three-dimensional molecular diagram information, wherein the three-dimensional molecular diagram characteristics specifically comprise:
determining the attention weight corresponding to the specified compound molecule through the graph attention mechanism network;
according to the attention weight and each embedded characteristic determined by the graph annotation mechanism network based on the three-dimensional molecular graph information, invariant characteristics and isovariable characteristics corresponding to the specified compound molecules are determined;
and determining the three-dimensional molecular map features according to the invariant features and the isovariable features.
The specification provides a task execution method based on molecular energy information, which comprises the following steps:
receiving a task execution request aiming at an original compound, and acquiring characterization data of molecules of the original compound;
processing the characterization data of the original compound molecules to determine three-dimensional molecular map information corresponding to the original compound molecules;
Inputting the three-dimensional molecular map information of the original compound molecules into a pre-trained prediction model to determine three-dimensional molecular map features corresponding to the original compound molecules through the prediction model, and determining molecular energy information corresponding to the original compound molecules according to the three-dimensional molecular map features corresponding to the original compound molecules, wherein the prediction model is trained by the model training method;
and executing tasks according to the molecular energy information corresponding to the original compound molecules.
Optionally, the tasks include: molecular regulation tasks or molecular design tasks;
executing tasks according to the molecular energy information corresponding to the original compound molecules, wherein the tasks specifically comprise:
determining the corresponding relation between the molecular structure and the molecular energy of the original compound molecules according to the molecular energy information corresponding to the original compound molecules and the characterization data corresponding to the original compound molecules;
and executing the molecular design task or the molecular regulation task based on the corresponding relation.
The present specification provides a model training apparatus comprising:
the device comprises an acquisition module, a detection module and a display module, wherein the acquisition module is used for acquiring characterization data of a specified compound molecule, and the characterization data are used for characterizing the position information and attribute information of each atom in the specified compound molecule;
The determining module is used for processing the characterization data and determining three-dimensional molecular diagram information of the specified compound molecules;
the input module is used for inputting the three-dimensional molecular diagram information into a prediction model to be trained so as to determine the three-dimensional molecular diagram characteristics corresponding to the specified compound molecules according to the three-dimensional molecular diagram information based on the isomorphism between the position information and the embedded characteristics corresponding to the position information and the invariance between the attribute information and the embedded characteristics corresponding to the attribute information through the prediction model;
the prediction module is used for predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular graph characteristics;
and the training module is used for training the prediction model by taking the deviation between the minimized predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule as an optimization target.
The present specification provides a task execution device based on molecular energy information, comprising:
the receiving module is used for receiving a task execution request aiming at an original compound and acquiring characterization data of molecules of the original compound;
The construction module is used for processing the characterization data of the original compound molecules and determining three-dimensional molecular diagram information corresponding to the original compound molecules;
the determining module is used for inputting the three-dimensional molecular diagram information of the original compound molecules into a pre-trained prediction model so as to determine the three-dimensional molecular diagram characteristics corresponding to the original compound molecules through the prediction model and determine the molecular energy information corresponding to the original compound molecules according to the three-dimensional molecular diagram characteristics corresponding to the original compound molecules, wherein the prediction model is obtained by training through the model training method;
and the execution module is used for executing tasks according to the molecular energy information corresponding to the original compound molecules.
The present specification provides a computer readable storage medium storing a computer program which when executed by a processor implements the method of model training described above or a method of task execution based on molecular energy information.
The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of model training described above or a method of task execution based on molecular energy information when executing the program.
The above-mentioned at least one technical scheme that this specification adopted can reach following beneficial effect:
in the model training method provided by the specification, the characterization data of the specified compound molecules are obtained, and three-dimensional molecular diagram information of the specified compound molecules is constructed; inputting the three-dimensional molecular diagram information into a prediction model, and determining the three-dimensional molecular diagram characteristics corresponding to the specified compound molecules; predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular diagram characteristics; the model is trained with the objective of optimizing the minimized deviation between the predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule. In the subsequent process of predicting the energy of the whole molecule by the molecular structure characteristics, the whole energy of the compound molecule under any structure can be rapidly and accurately predicted by the prediction model, and the accuracy of molecular energy prediction is improved.
According to the method, three-dimensional molecular diagram information of the compound molecules can be constructed through the characterization data of the compound molecules, the structure and the attribute of each atom in the compound molecules can be fully characterized, after the three-dimensional molecular diagram is input into a prediction model, characteristic extraction can be carried out based on isodegeneration of position information and invariance of attribute information, so that the characteristics of the compound molecules are accurately expressed, the original characteristics of the compound molecules on the attribute and the structure are fully ensured, then the energy information of the compound molecules is predicted according to the extracted characteristics, the model is trained according to the prediction result, the accuracy of the energy information of the compound molecules predicted by the prediction model after the training is fully improved, and effective guarantee is provided for a downstream molecular research task.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification, illustrate and explain the exemplary embodiments of the present specification and their description, are not intended to limit the specification unduly. In the drawings:
FIG. 1 is a flow chart of a method of model training provided in the present specification;
FIG. 2 is a schematic flow chart of a task execution method based on molecular energy information provided in the present specification;
FIG. 3 is a schematic diagram of the architecture of an exploration system for molecular structure-energy relationship of a compound provided in the present specification;
FIG. 4 is a schematic diagram of a model training apparatus provided in the present specification;
FIG. 5 is a schematic diagram of a task execution device based on molecular energy information provided in the present specification;
fig. 6 is a schematic structural diagram of an electronic device corresponding to fig. 1 or fig. 2 provided in the present specification.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the present specification more apparent, the technical solutions of the present specification will be clearly and completely described below with reference to specific embodiments of the present specification and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present specification. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.
The following describes in detail the technical solutions provided by the embodiments of the present specification with reference to the accompanying drawings.
Fig. 1 is a flow chart of a method for model training provided in the present specification, including the following steps:
s101: and obtaining characterization data of the specified compound molecules, wherein the characterization data are used for characterizing the position information and attribute information of each atom in the specified compound molecules.
S102: and processing the characterization data to determine three-dimensional molecular map information of the specified compound molecules.
The execution subject of the model training method provided in the present specification may be a terminal device such as a desktop computer or a notebook computer, and of course, may also be a server, and for convenience of explanation, the model training method provided will be described below using only the terminal device as an execution subject.
In this specification, the terminal device may obtain raw data of structures of various specified compound molecules to construct a data set for training a model, where the data may be obtained by crawling from an external network, or may be obtained in a file entry form.
After the data set is obtained, the terminal device needs to search out data suitable for training the prediction model, that is, the various specified compound molecules recorded in the data set are not all suitable for training as samples, some data may not have good tag molecular energy information, and some data may belong to "dirty data".
Therefore, the terminal device needs to search out the specified compound molecules suitable as training samples from the data set, and further determine the initial data of the specified compound molecules. The specific implementation manner can be that the created expandable three-dimensional molecular diagram structure data generator cleans, reconstructs and optimizes the class-specific compound molecular compound data, so as to search out the specific compound molecules serving as training samples and determine the three-dimensional molecular diagram information of the specific compound molecules.
In this specification, after determining initial data of a specified compound molecule from the data set, the terminal device may further determine characterization data of the specified compound molecule, where the characterization data is used to characterize position information and attribute information of each atom in the specified compound molecule, where the position information includes: the coordinates of each atom in the specified compound molecule under the specified coordinate system are specified, and the attribute information comprises: specifying the type of each atom in the compound molecule, specifying the directional vector between any two atoms in the compound molecule, and specifying the connection information between any two atoms in the compound molecule. Of course, other characterization data, such as chemical formula, molecular weight, chemical bond, structure, physical property, etc., may be included, which is not specifically limited in this specification.
After determining the characterization data of the specified compound molecule, the terminal device may reconstruct and optimize a part of the characterization data (including the attribute information and the position information), so as to obtain map data corresponding to the three-dimensional molecular map information of the specified compound molecule.
Specifically, the coordinates of each atom in a given compound molecule under a given coordinate system can be determined with reference to international standards and the class vector of each atom in the given compound molecule is encoded by a vector in one-hot format.
And the direction vector between any two atoms in the compound molecule and the connection information between any two atoms in the specified compound molecule can be represented as matrix data.
The vector of the direction between every two atoms of the molecule of the specified compound can be expressed as a vector with the direction obtained by the full connection mode of all atoms in the molecule, and the vector dimension can be 3*N (N=)Where n is the number of atoms in the molecule). The terminal device may define different connection information according to different atoms, for example, connection information between two atoms A-A may be represented by "0", and connection information between a-B may be represented by "1".
The distance between atoms used to characterize a given compound molecule and the distance unit vector can be determined by the following formula:
。
From the above, it can be seen that, in the present specification, the attribute and the position of each atom in the specified compound molecule are comprehensively considered in determining the three-dimensional molecular diagram information of the specified compound molecule, so that the finally determined three-dimensional molecular diagram information can comprehensively represent the characteristics of the specified compound molecule, and further the accuracy and the rationality of the subsequent prediction model on the prediction result are ensured.
S103: inputting the three-dimensional molecular map into a prediction model to be trained, and determining the three-dimensional molecular map features corresponding to the specified compound molecules according to the molecular map information of the three-dimensional molecular map based on the isomorphism between the embedded features corresponding to the position information and the invariance between the embedded features corresponding to the attribute information and the attribute information through the prediction model.
S104: and predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular graph characteristics.
The terminal equipment can input the three-dimensional molecular diagram of the specified compound molecule into a prediction model to be trained, so that the three-dimensional molecular diagram characteristic corresponding to the specified compound molecule is determined according to the molecular diagram information of the three-dimensional molecular diagram based on the isomorphism between the embedded characteristics corresponding to the position information and the invariance between the embedded characteristics corresponding to the attribute information through the prediction model and further according to the embedded characteristics.
Peer-to-peer denaturation and the invariance are specified as follows:
for functionsWherein x->X，f (x)/>Y, there is one variant G such as E (3) acting on X and Y, if for all X +.>X and all g->G, having f (G (x))=g (f (x)), the function f is said to have isovariability with respect to G, if for all x +.>X and all g->G, there is f (G (x))=f (x), then the function f is said to have invariance with respect to G.
In the present specification, the above-mentioned prediction model may be provided with a graph-annotation mechanism network and a multi-layer sensor constructed based on three-dimensional isodegeneration, and the process of the prediction model to finally predict the energy information of the specified compound molecule according to the inputted three-dimensional molecular graph information may be regarded as a graph embedding-feature regression process in practice.
Specifically, after the terminal device inputs the three-dimensional molecular diagram information of the specified compound molecules into the prediction model to be trained, the diagram annotation mechanism network can determine the atomic types and coordinate information in the compound molecules and the embedded features corresponding to the direction vectors and connection information among the atoms according to the three-dimensional molecular diagram information, and then determine the invariant features and the isovariable features corresponding to the compound molecules in the embedded features, and further determine the three-dimensional molecular diagram features of the specified compound molecules according to the invariant features and the isovariable features.
The terminal device can input the three-dimensional molecular diagram characteristics into the multi-layer sensor, and the multi-layer sensor is used for determining the energy information corresponding to the specified compound molecules.
Wherein, when determining the embedded features (including atomic coordinates and types, the embedded features corresponding to the direction vector and the connection information between every two atoms) of the molecular diagram of the specific compound, the determination can be performed by the following ways:
where a and b represent different atoms, f is an embedded eigenvalue, C is a Clebsch-gorman coefficient, Y is a real spherical harmonic, and R is a learnable function with respect to distance. The embedding mode of the molecules can enable the embedding characteristics to have the variability of SE (3) and the like for atomic coordinates, and the invariance of SE (3) can be met for the atomic types, the direction vector between any two atoms, the connection information and the like.
Of course, in practical applications, the embedding feature may be determined by other manners that can be implemented, which is not illustrated in the present specification.
After the above-mentioned embedded features are obtained, the embedded features can be further processed through an attention mechanism in practice to obtain more accurate three-dimensional molecular map features. Specifically, the graph attention mechanism network may determine the attention weight for the specified compound molecule, and then determine the invariant feature and the isovariable feature for the specified compound molecule according to the attention weight and the determined embedding feature, thereby determining the three-dimensional molecular graph feature of the specified compound molecule according to the invariant feature and the isovariable feature.
The above-described procedure for determining the three-dimensional molecular map features of a given compound molecule by the mechanism of attention will be described below with specific examples.
The embedded forms of query (Q) and key (K) can be obtained through a three-dimensional molecular graph attention mechanism as follows:
wherein,q characteristic for representing atom a in a given compound molecule,/->For representing specializationK characteristic of atom a≡b in compound molecule. Wherein (1)>Is->Abbreviation expression, meta-L>Is thatIs expressed by abbreviations of (a). At this time, the attention function is expressed as:
the addition to the molecular graph embedding formula draws attention mechanisms.
Further, the terminal device can predict the molecular energy information corresponding to the specified compound molecules through predicting the pooling function of the network and the multi-layer perceptron.
And training the prediction model with the deviation between the minimized predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule as an optimization target, wherein the process can be expressed as:
argmin{sum(∆E=|E p -E t |)}
therefore, in the process of predicting the molecular structure, the molecular energy can be rapidly and accurately predicted through the prediction model, so that the efficiency and the accuracy of exploring the constructed molecular structure-energy relation are improved.
In this specification, three-dimensional molecular map features of a given compound molecule can be determined by a map attention mechanism. As mentioned above, the process of predicting energy from the three-dimensional molecular map features of the specified compound molecule can be regarded as a process in which the multi-layer sensor predicts a new atom or a new molecular structure capable of being connected to the specified compound molecule from the invariant feature and the isovariable feature of each atom in the specified compound molecule, so that, when the three-dimensional molecular map features are determined, the updated invariant feature and the isovariable feature of each atom in the specified compound molecule can be obtained through the connection of the sensor in practice, and the prediction of energy can be performed.
In this specification, the prediction model may be actually trained by a joint training method. That is, after determining the molecular coordinate embedding characteristics of the compound, the neural network and the multilayer perceptron of the attention mechanism of the denaturation graph are adjusted together with the minimum loss value as an optimization target. Wherein the loss value can be practically determined by the deviation between the target predicted energy information and the actual energy information corresponding to the specified compound molecule.
After the prediction model is trained, the molecular structure information can be predicted through the trained prediction model, so that the recommendation of the molecular structure information is realized. The specific process is shown in the following figure.
Fig. 2 is a schematic flow chart of a task execution method based on molecular energy information provided in the present specification, including the following steps:
s201: a task execution request for an original compound is received and characterization data of molecules of the original compound is obtained.
S202: and processing the characterization data of the original compound molecules to determine three-dimensional molecular map information corresponding to the original compound molecules.
When the predictive model meets the training objective (e.g., reaches a preset number of training sessions or converges to a preset range), the terminal device may deploy it for performing downstream tasks.
In this specification, after receiving a task execution request for an original compound, a terminal device may acquire a prediction instruction of a molecular structure input by a user, so as to acquire characterization data of an originally specified compound molecule according to the prediction instruction.
And then carrying out reconstruction, optimization and other processing on the characterization data to obtain three-dimensional molecular diagram information of the specified compound molecules, wherein the determination of the three-dimensional molecular diagram information is basically consistent with the determination process of the three-dimensional molecular diagram information in the model training, and detailed description is omitted. The terminal device mentioned here may refer to a desktop computer, a notebook computer, etc.
S203: inputting the three-dimensional molecular map information of the original compound molecules into a pre-trained prediction model to determine three-dimensional molecular map features corresponding to the original compound molecules through the prediction model, and determining molecular energy information corresponding to the original compound molecules according to the three-dimensional molecular map features corresponding to the original compound molecules, wherein the prediction model is trained by the model training method.
The terminal device can input the three-dimensional molecular diagram information of the original specified compound molecules into a prediction model deployed in the terminal device, and the prediction model outputs energy information of molecules which are combined with the original specified compound molecules into preset functions.
It should be noted that, since the prediction model is already trained in the above model training process by the supervised training method, the reinforcement learning model, that is, the final molecular energy information output by the prediction model, may not be used in the actual application process, and further screening is not required.
S204: and executing tasks according to the molecular energy information corresponding to the original compound molecules.
After determining the molecular energy information of the specified compound molecule, the terminal device may determine a correspondence between the molecular structure and the molecular energy of the original compound molecule according to the molecular energy information corresponding to the original compound molecule and the characterization data corresponding to the original compound molecule, and then execute a molecular design task or the molecular regulation task based on the correspondence.
Of course, the terminal device may also recommend the correspondence between the molecular structure and the molecular energy of the original compound molecule as the information to be recommended to the user.
Further, the present disclosure also provides a system for exploring the molecular structure-energy relationship of a compound, as shown in fig. 3.
Fig. 3 is a schematic diagram of the architecture of an exploration system of molecular structure-energy relationship of a compound provided in the present specification.
As can be seen from fig. 3, the system is mainly composed of the following parts:
and the storage subsystem is used for storing the data set and storing the molecular energy information predicted by the prediction model in practical application and given structure information.
And the control subsystem is used for predicting the compound molecular energy information according to the three-dimensional molecular diagram information of the originally specified compound molecules input into the subsystem.
The control subsystem comprises three units, namely a molecular feature extraction unit, a molecular energy prediction training unit and a molecular energy prediction output unit, which are sequentially used for obtaining three-dimensional molecular diagram features, molecular structure-prediction model training result information and obtaining molecular energy information according to new input molecular structure information.
Compared with the prior art, the prediction model in the application greatly improves the efficiency and accuracy of relation construction, and with the continuous development of fields such as artificial intelligence and machine learning and continuous penetration in the molecular science and technology direction, the performance prediction and design of molecules become more complex compared with the prior fully-connected network. Molecular structure-energy relation prediction models based on artificial intelligence or machine learning depend on a representation method of molecular characteristics, and molecular graph representation can be used for describing the structural characteristics of molecules by regarding intramolecular atoms and chemical bonds as nodes and edges. By combining the deep network model method, a developer can effectively construct the structure-energy relation of the compound molecules so as to provide powerful technical methods and means for researching the molecular energy related dynamics characteristics and designing downstream tasks such as a molecule with a certain function according to the requirements of users.
However, the currently adopted method cannot more comprehensively characterize molecules and cannot efficiently predict the molecular structure-energy relation information of the specific compounds.
Therefore, in the model training method provided by the specification, in the process of determining the three-dimensional molecular diagram information of the specified compound molecule, the characterization data of various molecular structures of the specified compound molecule are comprehensively referenced, so that the finally determined three-dimensional molecular diagram information can comprehensively characterize the molecular structure characteristics of the specified compound molecule.
In addition, when the three-dimensional molecular diagram characteristics of the specified compound molecules are determined, the characteristics of the specified compound molecules on the molecular structure can be fully represented by the method because the three-dimensional molecular diagram characteristics of the specified compound molecules are determined according to the invariant characteristics and the isovariable characteristics of the specified compound molecules, so that the prediction model can be used for accurately and reasonably predicting energy information through the three-dimensional molecular diagram characteristics of the specified compound molecules.
The foregoing is a schematic diagram of a corresponding model training device and a task execution device based on molecular energy information, which are based on the same thought, as shown in fig. 4 and 5.
Fig. 4 is a schematic diagram of a model training device provided in the present specification, including:
the obtaining module 401 is configured to obtain characterization data of a specified compound molecule, where the characterization data is used to characterize position information and attribute information of each atom in the specified compound molecule;
a determining module 402, configured to process the characterization data, and determine three-dimensional molecular map information of the specified compound molecule;
an input module 403, configured to input the three-dimensional molecular map information into a prediction model to be trained, so as to determine, according to the three-dimensional molecular map information, a three-dimensional molecular map feature corresponding to the specified compound molecule based on an isomorphism between the position information and the embedded feature corresponding to the position information and an invariance between the attribute information and the embedded feature corresponding to the attribute information, by using the prediction model;
A prediction module 404, configured to predict molecular energy information corresponding to the specified compound molecule according to the three-dimensional molecular graph feature;
and the training module 405 is configured to train the prediction model with a deviation between the minimized predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule as an optimization target.
Optionally, the obtaining module 401 is specifically configured to select initial data of the specified compound molecule from a data set of molecular compounds; determining the characterization data based on the initial data for the specified compound molecule.
Optionally, the location information includes: coordinates of each atom in the molecule of the specified compound under a specified coordinate system;
the attribute information includes: the type of each atom in the molecule of the specified compound, the direction vector between any two atoms in the molecule of the specified compound and the connection information between any two atoms in the molecule of the specified compound.
Optionally, a graph annotation force mechanism network is arranged in the prediction model;
the input module 403 is specifically configured to determine, through the graph attention mechanism network, an attention weight corresponding to the specified compound molecule; according to the attention weight and each embedded characteristic determined by the graph annotation mechanism network based on the three-dimensional molecular graph information, invariant characteristics and isovariable characteristics corresponding to the specified compound molecules are determined; and determining the three-dimensional molecular map features according to the invariant features and the isovariable features.
Fig. 5 is a schematic diagram of a task execution device based on molecular energy information provided in the present specification, including:
a receiving module 501, configured to receive a task execution request for an original compound, and obtain characterization data of a molecule of the original compound;
the construction module 502 is configured to process the characterization data of the original compound molecule, and determine three-dimensional molecular map information corresponding to the original compound molecule;
a determining module 503, configured to input three-dimensional molecular map information of the original compound molecule into a pre-trained prediction model, so as to determine, according to the prediction model, three-dimensional molecular map features corresponding to the original compound molecule, and determine molecular energy information corresponding to the original compound molecule according to the three-dimensional molecular map features corresponding to the original compound molecule, where the prediction model is obtained by training by using the model training method described above;
and the execution module 504 is used for executing tasks according to the molecular energy information corresponding to the original compound molecules.
Optionally, the tasks include: molecular regulation tasks or molecular design tasks;
the execution module 504 is specifically configured to determine a correspondence between a molecular structure and molecular energy of the original compound molecule according to the molecular energy information corresponding to the original compound molecule and the characterization data corresponding to the original compound molecule; and executing the molecular design task or the molecular regulation task based on the corresponding relation.
The present disclosure also provides a computer readable storage medium storing a computer program operable to perform a model training method as provided in fig. 1 or a task execution method based on molecular energy information as provided in fig. 2.
The present specification also provides a schematic structural diagram of an electronic device corresponding to fig. 1 or fig. 2 shown in fig. 6. At the hardware level, as shown in fig. 6, the electronic device includes a processor, an internal bus, a network interface, a memory, and a nonvolatile storage, and may of course include hardware required by other services. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to implement the model training method described in fig. 1 or the molecular structure information recommendation method described in fig. 2.
The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functions of each element may be implemented in one or more software and/or hardware elements when implemented in the present specification.
It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The present description is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments.
The foregoing is merely exemplary of the present disclosure and is not intended to limit the disclosure. Various modifications and alterations to this specification will become apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, or the like, which are within the spirit and principles of the present description, are intended to be included within the scope of the claims of the present description.
Claims (10)
1. A method of model training, comprising:
acquiring characterization data of a specified compound molecule, wherein the characterization data are used for characterizing position information and attribute information of each atom in the specified compound molecule;
processing the characterization data to determine three-dimensional molecular map information of the specified compound molecules;
inputting the three-dimensional molecular diagram information into a prediction model to be trained, and determining three-dimensional molecular diagram characteristics corresponding to the specified compound molecules according to the three-dimensional molecular diagram information based on isomorphism between the position information and the embedded characteristics corresponding to the position information and invariance between the attribute information and the embedded characteristics corresponding to the attribute information through the prediction model;
predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular graph characteristics;
and training the prediction model by taking the deviation between the predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule as an optimization target.
2. The method of claim 1, wherein obtaining characterization data for a given compound molecule, specifically comprises:
Selecting initial data for the specified compound molecule in a dataset of molecular compounds;
determining the characterization data based on the initial data for the specified compound molecule.
3. The method of claim 1, wherein the location information comprises: coordinates of each atom in the molecule of the specified compound under a specified coordinate system;
the attribute information includes: the type of each atom in the molecule of the specified compound, the direction vector between any two atoms in the molecule of the specified compound and the connection information between any two atoms in the molecule of the specified compound.
4. The method of claim 1, wherein a graph annotation mechanism network is provided in the predictive model;
determining the three-dimensional molecular diagram characteristics corresponding to the specified compound molecules according to the three-dimensional molecular diagram information, wherein the three-dimensional molecular diagram characteristics specifically comprise:
determining the attention weight corresponding to the specified compound molecule through the graph attention mechanism network;
according to the attention weight and each embedded characteristic determined by the graph annotation mechanism network based on the three-dimensional molecular graph information, invariant characteristics and isovariable characteristics corresponding to the specified compound molecules are determined;
And determining the three-dimensional molecular map features according to the invariant features and the isovariable features.
5. A method for performing a task based on molecular energy information, comprising:
receiving a task execution request aiming at an original compound, and acquiring characterization data of molecules of the original compound;
processing the characterization data of the original compound molecules to determine three-dimensional molecular map information corresponding to the original compound molecules;
inputting the three-dimensional molecular map information of the original compound molecules into a pre-trained prediction model to determine three-dimensional molecular map features corresponding to the original compound molecules through the prediction model, and determining molecular energy information corresponding to the original compound molecules according to the three-dimensional molecular map features corresponding to the original compound molecules, wherein the prediction model is obtained by training by the method according to any one of claims 1-4;
and executing tasks according to the molecular energy information corresponding to the original compound molecules.
6. The method of claim 5, wherein the task comprises: molecular regulation tasks or molecular design tasks;
Executing tasks according to the molecular energy information corresponding to the original compound molecules, wherein the tasks specifically comprise:
determining the corresponding relation between the molecular structure and the molecular energy of the original compound molecules according to the molecular energy information corresponding to the original compound molecules and the characterization data corresponding to the original compound molecules;
and executing the molecular design task or the molecular regulation task based on the corresponding relation.
7. A model training device, comprising:
the device comprises an acquisition module, a detection module and a display module, wherein the acquisition module is used for acquiring characterization data of a specified compound molecule, and the characterization data are used for characterizing the position information and attribute information of each atom in the specified compound molecule;
the determining module is used for processing the characterization data and determining three-dimensional molecular diagram information of the specified compound molecules;
the input module is used for inputting the three-dimensional molecular diagram information into a prediction model to be trained so as to determine the three-dimensional molecular diagram characteristics corresponding to the specified compound molecules according to the three-dimensional molecular diagram information based on the isomorphism between the position information and the embedded characteristics corresponding to the position information and the invariance between the attribute information and the embedded characteristics corresponding to the attribute information through the prediction model;
The prediction module is used for predicting molecular energy information corresponding to the specified compound molecules according to the three-dimensional molecular graph characteristics;
and the training module is used for training the prediction model by taking the deviation between the minimized predicted molecular energy information and the actual molecular energy information corresponding to the specified compound molecule as an optimization target.
8. A task execution device based on molecular energy information, comprising:
the receiving module is used for receiving a task execution request aiming at an original compound and acquiring characterization data of molecules of the original compound;
the construction module is used for processing the characterization data of the original compound molecules and determining three-dimensional molecular diagram information corresponding to the original compound molecules;
a determining module, configured to input three-dimensional molecular map information of the original compound molecule into a pre-trained prediction model, so as to determine, according to the prediction model, a three-dimensional molecular map feature corresponding to the original compound molecule, and determine, according to the three-dimensional molecular map feature corresponding to the original compound molecule, molecular energy information corresponding to the original compound molecule, where the prediction model is obtained by training by a method according to any one of claims 1 to 4;
And the execution module is used for executing tasks according to the molecular energy information corresponding to the original compound molecules.
9. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-6.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of the preceding claims 1-6 when executing the program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311702625.8A CN117393075A (en) | 2023-12-12 | 2023-12-12 | Model training method and task execution method based on molecular energy information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311702625.8A CN117393075A (en) | 2023-12-12 | 2023-12-12 | Model training method and task execution method based on molecular energy information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117393075A true CN117393075A (en) | 2024-01-12 |
Family
ID=89463540
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311702625.8A Pending CN117393075A (en) | 2023-12-12 | 2023-12-12 | Model training method and task execution method based on molecular energy information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117393075A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116524998A (en) * | 2023-06-15 | 2023-08-01 | 之江实验室 | Model training method and molecular property information prediction method and device |
| CN116597892A (en) * | 2023-05-15 | 2023-08-15 | 之江实验室 | Model training method and molecular structure information recommending method and device |
| CN116798536A (en) * | 2023-06-12 | 2023-09-22 | 杭州碳硅智慧科技发展有限公司 | Training method and device for molecular multi-conformation prediction model |
| CN116959605A (en) * | 2023-01-12 | 2023-10-27 | 腾讯科技（深圳）有限公司 | Molecular property prediction method, training method and device of molecular property prediction model |
-
2023
- 2023-12-12 CN CN202311702625.8A patent/CN117393075A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116959605A (en) * | 2023-01-12 | 2023-10-27 | 腾讯科技（深圳）有限公司 | Molecular property prediction method, training method and device of molecular property prediction model |
| CN116597892A (en) * | 2023-05-15 | 2023-08-15 | 之江实验室 | Model training method and molecular structure information recommending method and device |
| CN116798536A (en) * | 2023-06-12 | 2023-09-22 | 杭州碳硅智慧科技发展有限公司 | Training method and device for molecular multi-conformation prediction model |
| CN116524998A (en) * | 2023-06-15 | 2023-08-01 | 之江实验室 | Model training method and molecular property information prediction method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wiewel et al. | Latent space physics: Towards learning the temporal evolution of fluid flow | |
| EP3579169A1 (en) | Learned model provision method, and learned model provision device | |
| CN115456160A (en) | Data processing method and data processing equipment | |
| CN113688304A (en) | Training method for search recommendation model, and method and device for sequencing search results | |
| CN111581414B (en) | Method, device, equipment and storage medium for identifying, classifying and searching clothes | |
| CN111651671B (en) | User object recommendation method, device, computer equipment and storage medium | |
| CN111782838A (en) | Image question-answering method, image question-answering device, computer equipment and medium | |
| CN111950596A (en) | Training method for neural network and related equipment | |
| CN116249991A (en) | Neural network distillation method and device | |
| KR20190056940A (en) | Method and device for learning multimodal data | |
| CN113515589B (en) | Data recommendation method, device, equipment and medium | |
| CN115238909A (en) | Data value evaluation method based on federal learning and related equipment thereof | |
| CN116109449A (en) | Data processing method and related equipment | |
| CN114861859A (en) | Training method of neural network model, data processing method and device | |
| CN115879508A (en) | Data processing method and related device | |
| CN111260074B (en) | Method for determining hyper-parameters, related device, equipment and storage medium | |
| CN112052681A (en) | Information extraction model training method, information extraction device and electronic equipment | |
| CN112232360A (en) | Image retrieval model optimization method, image retrieval device and storage medium | |
| KR102561799B1 (en) | Method and system for predicting latency of deep learning model in device | |
| CN117393075A (en) | Model training method and task execution method based on molecular energy information | |
| WO2023149838A2 (en) | Machine learning with periodic data | |
| CN115795025A (en) | Abstract generation method and related equipment thereof | |
| CN115907041A (en) | Model training method and device | |
| CN114707070A (en) | User behavior prediction method and related equipment thereof | |
| Borowik et al. | Computational Intelligence and Efficiency in Engineering Systems |
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 |