Material design method and device based on electronic structure, equipment and storage medium

Abstract

The invention relates to the technical field of new material design, in particular to a material design method and device based on an electronic structure, equipment and a storage medium. According to the material design method based on the electronic structure, a quantum chemistry calculation method and a front line molecular orbital theory are introduced, and a molecular generation model is constructed according to the interaction relationship of groups in molecules, so that predicted new material molecules follow required electronic structure characteristics, and the method is more suitable for design, research and development in the field of photoelectric materials.

Description

Technical Field

[0001] The present invention relates to the technical field of new material design, and in particular to a material design method, device, equipment and storage medium based on electronic structure. Background Art

[0002] When conducting high-throughput screening in materials science, rapidly designing a large number of candidate molecules with target properties is a crucial step. Currently, through the close integration of computational simulation, model prediction and experimental verification, a complete closed loop of new material exploration can be formed, thereby achieving rapid iteration and discovery of high-performance materials. Generating stable and efficient candidate molecular structures can not only significantly reduce the resource investment required for computational simulation, model prediction and experimental verification, but also reduce the complexity of new material development. Therefore, efficient and accurate molecular design methods play a vital role in accelerating material innovation and optimizing R&D processes.

[0003] At present, artificial intelligence methods have greatly promoted the research and development of new materials, especially in the generation of candidate molecules and the prediction of key properties of materials. In the process of AI-assisted new material molecular design, generative models are usually used to generate candidate molecules. Deep learning models such as language model CLM, autoencoder model VAE, adversarial generation model GAN, diffusion model, etc. have been proven to be able to reliably generate a large number of new organic molecules. When generating new molecules, the usual method is to fine-tune the pre-trained model using molecules with target properties. A few methods generate molecules by directly inputting candidate molecule properties or groups into the model as conditional constraints. The traditional candidate molecule generation method based on the above method only uses the learning of geometric structure information as the basis for the design of new molecules, which can generally meet the needs in the design of biomedical molecules. However, when generating candidate molecules with complex electronic structures such as optoelectronic materials, the electronic structure of the candidate molecules must meet specific conditions, otherwise the target properties cannot be exhibited. The traditional molecule generation method only focuses on the geometric structure and ignores the electronic structure, which is often difficult to be competent for the design and development of optoelectronic materials.

[0004] It can be seen that the existing technology still needs to be improved and enhanced. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a material design method, device, equipment and storage medium based on electronic structure, aiming to solve the problem that the existing molecular design methods are not suitable for designing optoelectronic materials with complex electronic structures.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The first aspect of the present invention provides a material design method based on electronic structure, comprising the following steps: obtaining multiple existing molecular structures, using quantum chemical calculation methods to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set; based on the intramolecular group interaction relationship, constructing a cross-modal information extraction model according to the molecular structure information data set; obtaining the three-dimensional geometric structure of the target molecule, using the cross-modal information extraction model, calculating the cross-modal information of the target molecule according to the three-dimensional geometric structure of the target molecule, and integrating the obtained cross-modal information into the molecular structure information data set; constructing a molecular generation model according to the molecular structure information data set, obtaining the molecular information of the target material to be predicted, and using the molecular generation model to predict according to the molecular information of the target material to be predicted to obtain the target material molecule.

[0008] Optionally, in a first implementation of the first aspect of the present invention, the obtaining of multiple existing molecular structures, and the use of quantum chemical calculation methods to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set specifically include: obtaining three-dimensional coordinate structure information of multiple organic molecules, and using quantum chemical calculation methods to calculate the charge distribution in the frontier molecular orbitals for each organic molecule, and establishing a label for each atom of the organic molecule according to the calculation results to obtain a molecular structure information data set.

[0009] Optionally, in a second implementation method of the first aspect of the present invention, the cross-modal information extraction model is constructed based on the interaction relationship between intramolecular groups and the molecular structure information data set, specifically including: setting a first module, the first module is used to extract the cumulative characteristics of each atom in the three-dimensional geometric structure of the molecule; setting a second module, the second module is used to calculate the electronic structure information of each atom based on the cumulative characteristics of each atom; integrating the first module and the second module to obtain a basic model; using the molecular structure information data set to verify the prediction results of the basic model, and optimizing the basic model according to the verification results to obtain a cross-modal information extraction model.

[0010] Optionally, in a third implementation of the first aspect of the present invention, the first module is set, and the first module is used to extract the cumulative characteristics of each atom in the three-dimensional geometric structure of the molecule, specifically including: setting a multi-level sampling structure, for obtaining element information and multi-level collection sampling results based on the three-dimensional geometric structure of the molecule; setting a geometric structure feature extraction model, for calculating multi-level atomic features based on the element information and multi-level collection sampling results; setting a multi-level feature accumulation model, for calculating the cumulative characteristics of each atom based on the multi-level atomic features.

[0011] Optionally, in a fourth implementation of the first aspect of the present invention, the second module is set, and the second module is used to calculate the electronic structure information of each atom according to the cumulative characteristics of each atom, specifically including: setting an inter-group attention interaction sub-model, which is used to calculate the global interaction characteristics of each atom according to the cumulative characteristics of each atom; setting an electronic structure distribution output model, which is used to calculate the electronic structure prediction result of each atom according to the global interaction characteristics of each atom.

[0012] Optionally, in a fifth implementation of the first aspect of the present invention, the three-dimensional geometric structure of the target molecule is obtained, a cross-modal information extraction model is used, the cross-modal information of the target molecule is calculated according to the three-dimensional geometric structure of the target molecule, and the obtained cross-modal information is integrated into the molecular structure information dataset, specifically including: generating multiple electron distribution prediction models for different frontier molecular orbitals based on the cross-modal information extraction model; obtaining the three-dimensional geometric structure of the target molecule, using multiple electron distribution prediction models for different frontier molecular orbitals, and predicting according to the three-dimensional geometric structure of the target molecule to obtain multiple frontier molecular orbital charge distribution prediction values ​​for each atom; sorting the multiple frontier molecular orbital charge distribution prediction values ​​according to the numerical values, and taking one or more atoms with the largest numerical values ​​as the central distribution atoms of the frontier molecular orbitals of the target molecule; extracting the global interaction features corresponding to the central distribution atoms, merging the frontier molecular orbital charge distribution prediction values ​​of the central distribution atoms with the global interaction features to obtain cross-modal information, and integrating the obtained cross-modal information into the molecular structure information dataset.

[0013] Optionally, in a sixth implementation of the first aspect of the present invention, the molecular generation model is constructed based on the molecular structure information data set, the molecular information of the target material to be predicted is obtained, and the molecular generation model is used to predict the target material according to the molecular information to be predicted, so as to obtain the target material molecules, specifically including: constructing an initial model using the molecular structure information data set and the molecular information of known materials, wherein the molecular information of the known materials includes molecular skeletons, group structures or physical and chemical properties, and the initial model is used to generate predicted molecular descriptors; formatting the molecular three-dimensional geometric structure corresponding to the molecular structure information data set to obtain a target molecular descriptor, calculating the loss value between the target molecular descriptor and the predicted molecular descriptor, and using the loss value to train the initial model to obtain a molecular generation model; obtaining the molecular information of the target material to be predicted, and using the molecular generation model to predict the target material according to the molecular information to be predicted, so as to obtain the target material molecules.

[0014] The second aspect of the present invention provides an electronic structure-based material design device, including: a first calculation module, used to obtain multiple existing molecular structures, and use quantum chemical calculation methods to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set; a first construction module, used to construct a cross-modal information extraction model based on the molecular structure information data set based on the intramolecular group interaction relationship; a second calculation module, used to obtain the three-dimensional geometric structure of the target molecule, use the cross-modal information extraction model, calculate the cross-modal information of the target molecule based on the three-dimensional geometric structure of the target molecule, and integrate the obtained cross-modal information into the molecular structure information data set; a second construction module, used to construct a molecular generation model based on the molecular structure information data set, obtain the molecular information of the target material to be predicted, and use the molecular generation model to predict according to the molecular information of the target material to be predicted to obtain the target material molecule.

[0015] The third aspect of the present invention provides an electronic structure-based material design device, comprising a memory and at least one processor, wherein the memory stores computer-readable instructions; the at least one processor calls the computer-readable instructions in the memory to execute the various steps of the electronic structure-based material design method as described above.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium having computer-readable instructions stored thereon. When the computer-readable instructions are executed by a processor, the various steps of the electronic structure-based material design method described above are implemented.

[0017] Beneficial effects: The present invention provides a material design method based on electronic structure, which first obtains multiple existing molecular structures, adopts a quantum chemical calculation method, and calculates the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set; then, based on the intramolecular group interaction relationship, a cross-modal information extraction model is constructed according to the molecular structure information data set and a molecular structure information data set is further obtained; by introducing the frontier molecular orbital theory, higher explainability and theoretical basis are provided for the generation of new molecules and the iterative optimization of the model; finally, a molecular generation model is constructed according to the molecular structure information data set for predicting new material molecules, ensuring that the generated new molecules follow the required electronic structure characteristics during the generation process, thereby having the expected target properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A first flow chart of the express information leakage prevention method provided in an embodiment of the present invention.

[0019] Figure 2 A second flow chart of the express information leakage prevention method provided in an embodiment of the present invention.

[0020] Figure 3 Schematic diagram of the structure of the cross-modal information extraction model.

[0021] Figure 4 A third flow chart of the express information leakage prevention method provided in an embodiment of the present invention.

[0022] Figure 5 A fourth flow chart of the express information leakage prevention method provided in an embodiment of the present invention.

[0023] Figure 6 A fifth flow chart of the express information leakage prevention method provided in an embodiment of the present invention.

[0024] Figure 7 The process of the application example of the present invention Figure 1 .

[0025] Figure 8 The process of the application example of the present invention Figure 2 .

[0026] Fig. 9 The process of the application example of the present invention Figure 3 .

[0027] Fig.10 A schematic diagram of the structure of an express information leakage prevention device provided in an embodiment of the present invention.

[0028] Fig.11 A schematic diagram of the structure of an express information leakage prevention device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention provides a material design method, device, equipment and storage medium based on electronic structure. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0030] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , the first embodiment of the electronic structure-based material design method in the embodiment of the present invention includes:

[0032] S101. Obtain multiple existing molecular structures, and use quantum chemical calculation methods to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set;

[0033] This step mainly uses quantum chemical calculations to calculate the electron distribution of each frontier molecular orbital based on the existing molecular structure, which serves as the data basis for the next step of model establishment.

[0034] S102. Based on the intramolecular group interaction relationship, a cross-modal information extraction model is constructed according to the molecular structure information dataset;

[0035] By introducing the frontier molecular orbital theory in the process of constructing the cross-modal information extraction model, it can provide higher explainability and theoretical basis for the generation of new molecules and the iterative optimization of the model. On this basis, the cross-modal information extraction model is further constructed based on the interaction relationship between intramolecular groups, which can improve the accuracy of model prediction. In addition, by setting up a cross-modal information extraction model, the acquisition of cross-modal information of electronic structure can be accelerated and efficiency can be improved.

[0036] S103. Obtaining the three-dimensional geometric structure of the target molecule, using a cross-modal information extraction model, calculating the cross-modal information of the target molecule according to the three-dimensional geometric structure of the target molecule, and integrating the obtained cross-modal information into a molecular structure information dataset;

[0037] S104. Construct a molecular generation model based on the molecular structure information data set to obtain the molecular information of the target material to be predicted, and use the molecular generation model to perform prediction based on the molecular information of the target material to be predicted to obtain the target material molecule.

[0038] The molecular structure information data set includes electronic structure characteristics, which are used as conditional constraints to establish and train the molecular generation model, ensuring that the generated new molecules follow the required electronic structure characteristics during the generation process, so as to have the expected target properties. After the generated molecules are verified for structure and synthesizability, a large number of new material molecules that meet the target electronic structure and target properties can be obtained.

[0039] In a second embodiment of the electronic structure-based material design method in an embodiment of the present invention, the acquisition of multiple existing molecular structures and the use of quantum chemical calculation methods to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set specifically include:

[0040] The three-dimensional coordinate structure information of multiple organic molecules is obtained, and the charge distribution in the frontier molecular orbital of each organic molecule is calculated using quantum chemical calculation methods. A label is established for each atom of the organic molecule based on the calculation results to obtain a molecular structure information data set.

[0041] When obtaining the three-dimensional coordinate structure information of organic molecules, in addition to the general organic molecules existing in nature, as many real molecules that exist in reality with similar principles or structures as the target generated molecules are included as much as possible. Based on the quantum chemical calculation method, the distribution of charges in the frontier molecular orbitals of each molecule is calculated, and the calculated distribution information is used to establish a label for each atom of the molecule, thereby obtaining a molecular structure information data set.

[0042] See also Figure 2 In a third embodiment of the electronic structure-based material design method in an embodiment of the present invention, the construction of a cross-modal information extraction model includes the following steps:

[0043] S201. Setting a first module, the first module is used to extract the cumulative features of each atom in the three-dimensional geometric structure of the molecule;

[0044] S202. Setting a second module, the second module is used to calculate the electronic structure information of each atom according to the accumulated characteristics of each atom;

[0045] S203. Integrate the first module and the second module to obtain a basic model;

[0046] S204. Use the molecular structure information dataset to verify the prediction results of the basic model, optimize the basic model according to the verification results, and obtain a cross-modal information extraction model.

[0047] In specific implementation, the cross-modal information extraction model includes two modules. The first module mainly performs multi-level sampling based on atoms, and extracts and integrates the characteristics of each atom at multi-level geometric structures; the second module mainly associates geometric structure characteristics and electronic structure characteristics based on inter-group interactions.

[0048] See also Figure 3 In the fourth embodiment of the electronic structure-based material design method according to the embodiment of the present invention, the setting of the first module (multi-level geometric feature extraction module A) includes the following steps:

[0049] S301. Setting a multi-level sampling structure for obtaining element information and multi-level collection sampling results according to the three-dimensional geometric structure of the molecule;

[0050] S302. Setting a geometric structure feature extraction model for calculating multi-level atomic features based on element information and multi-level set sampling results;

[0051] S303. Setting a multi-level feature accumulation model to calculate the cumulative features of each atom based on the multi-level atomic features.

[0052] In a specific embodiment, the first module includes a multi-level sampling structure Sampling, a geometric structure feature extraction model M geo and multi-level feature accumulation model M acc Both models can be built based on message passing neural networks (MPNNs), graph convolutions, or graph transformers.

[0053] For the input molecular 3D geometry X∈R N×3 The first module first applies a multi-level sampling structure Sampling to each atom i in k different geometric space ranges r k The structure contained in the sample is sampled to obtain the multi-level sampling result set {S k,i} and element information set {element k,i}.

[0054] The subscript k represents the different level numbers, and the sampling result of level k is determined by the geometric space range parameter r set in the multi-level sampling structure Sampling. k Decision. k The larger the value, the wider the sampling range and the larger the sampled geometry.

[0055] The sampling process not only obtains the multi-level sampling results S of the geometric structure around the sampling center atom i k , you can also get the element information element in these structures k,i .

[0056] Each atom i in the molecule is sampled in different ranges to obtain a multi-level sampling result set {S k,i} and element information set {element k,i}Afterwards, these sets are input into the geometric feature extraction model Mgeo to obtain the atomic features f at each level k,i , and further form multi-level atomic features. The multi-level atomic features of all atoms in the molecule form a set {f k,i}.

[0057] Multi-level atomic feature composition set {f k,i} Input the multi-level feature accumulation model, and finally obtain the accumulated feature h of each atom after calculation i .

[0058] Here is the formula:

[0059] S k,i ,element k,i =Sampling i (X,r k )

[0060] f k,i =M geo (S k,i ,element k,i )

[0061]

[0062] Refer to the above formula and use model M acc Perform operations on each atom i. Define the neighboring atom set N of each atom i i , for each neighbor set N of atom i i atom j, first based on the atomic features f of atoms i and j k,i and f k,j The nonlinear transformation σ is used to calculate the features that need to be accumulated at the k level, based on the distance l between atoms i and j. i，j , using σ d Calculate the coefficient of the cumulative feature, and then based on the coefficient a of each level k Accumulate multiple levels, this a k It can be designed empirically in advance or by using a set of atomic features {f k,i} Real-time calculation. Each atom calculates the accumulated characteristics h i Together they form the collection {h i}, used for calculation of the second module.

[0063] See also Figure 3In the fifth embodiment of the electronic structure-based material design method in the embodiment of the present invention, the setting of the second module (intergroup interaction mechanism module B) includes the following steps:

[0064] S401. Setting an inter-group attention interaction sub-model to calculate the global interaction feature of each atom according to the accumulated features of each atom;

[0065] S402. Setting an electronic structure distribution output model to calculate the electronic structure prediction result of each atom according to the global interaction characteristics of each atom.

[0066] Specifically, the second module includes the inter-group attention interaction sub-model M cross and electronic structure distribution output model M e The two models can be implemented using self-attention mechanism models, gated CNN\gated RNN and other methods involving mutual influence factor calculation modules.

[0067] Intergroup attention interaction model M cross and electronic structure distribution output model M e As shown in the following formula:

[0068] g i =M cross ({h i,i∈G})=σ a ({h i,i∈G})*h i +σ b ({h i,i∈G})

[0069]

[0070] Intergroup attention interaction model M cross Based on the accumulated features h of the current atom i i and the cumulative characteristic set {h i，i∈G}, using the nonlinear method σ a and σ b Get the cumulative characteristic h of the current atom i The scaling factor and bias, and then h i Calculate together to get the global interaction feature g of the current atom i i ;

[0071] Then the electronic structure distribution output model M e According to the global interaction feature g i Calculate the predicted electronic structure of the current atom Thus, the second module predicts the electronic structure information for each atom.

[0073] See also Figure 4 In the sixth embodiment of the electronic structure-based material design method in the embodiments of the present invention, the three-dimensional geometric structure of the target molecule is obtained, a cross-modal information extraction model is used, the cross-modal information of the target molecule is calculated according to the three-dimensional geometric structure of the target molecule, and the obtained cross-modal information is integrated into the molecular structure information data set, specifically including:

[0074] S501. Generate multiple electron distribution prediction models for different frontier molecular orbitals based on the cross-modal information extraction model;

[0075] Based on the cross-modal information extraction model of electronic structure established above, when studying the electronic structure of a specific frontier molecular orbital, multiple electron distribution prediction models can be established and trained to predict the electronic structure of different frontier molecular orbitals. j To predict the distribution;

[0076] S502. Obtaining the three-dimensional geometric structure of the target molecule, using a plurality of electron distribution prediction models for different frontier molecular orbitals, and performing predictions based on the three-dimensional geometric structure of the target molecule to obtain a plurality of frontier molecular orbital charge distribution prediction values ​​for each atom;

[0077] After training multiple frontier molecular orbital electron distribution prediction models respectively, the target frontier molecular orbital O j The cross-modal information of the electronic structure C j predictions;

[0078] S503. Sort the predicted values ​​of the frontier molecular orbital charge distribution according to their numerical values, and use one or more atoms with the largest numerical values ​​as the central distribution atoms of the frontier molecular orbital of the target molecule;

[0079] S504. Extract the global interaction features corresponding to the central distribution atoms, merge the frontier molecular orbital charge distribution prediction values ​​of the central distribution atoms with the global interaction features to obtain cross-modal information, and integrate the obtained cross-modal information into the molecular structure information dataset.

[0080] First, the three-dimensional geometric structure of the molecule is input into the target frontier molecular orbital electronic structure prediction model (A j , B j ), the model will give the predicted value of the frontier molecular orbital charge distribution of each atom, sort them according to the value, select one or more atoms with the largest value as the central distribution atom of the target frontier molecular orbital, and query model B by number at the same time j The atomic global interaction features in {g i} j, after merging, is obtained as the target molecular orbital O j The cross-modal information C j .

[0081] See also Figure 6 In the seventh embodiment of the electronic structure-based material design method according to the embodiments of the present invention, the molecular generation model is constructed according to the molecular structure information data set to obtain the molecular information of the target material to be predicted, and the molecular generation model is used to predict the target material according to the molecular information to be predicted to obtain the target material molecule, specifically including:

[0082] S601. Using a molecular structure information dataset and molecular information of known materials to construct an initial model, wherein the molecular information of the known materials includes a molecular skeleton, a group structure or a physicochemical property, and using the initial model to generate a predicted molecular descriptor;

[0083] S602. Convert the molecular three-dimensional geometric structure corresponding to the molecular structure information data set to obtain a target molecular descriptor, calculate the loss value between the target molecular descriptor and the predicted molecular descriptor, and use the loss value to train the initial model to obtain a molecular generation model;

[0084] When training the molecular generation model, all parameters of the cross-modal information extraction model are frozen, and it is only used to extract cross-modal information of the electronic structure of the input molecule.

[0085] S603. Acquire the molecular information of the target material to be predicted, and use the molecular generation model to perform prediction based on the molecular information of the target material to be predicted to obtain the target material molecule.

[0086] Specifically, first, for each organic molecule participating in the training, its three-dimensional binding structure X is input into the cross-modal information extraction model to obtain the cross-modal information set of the target molecular orbital {C j};

[0087] Then, with {C j} is used as a constraint, and together with other molecular generation conditions, it is used as the input constraint of the initial model. Other generation conditions here include but are not limited to geometric structure encoding such as molecular skeleton and group structure and some basic physical and chemical properties. The initial model can select the currently commonly used molecular prediction model with conditional constraints.

[0088] Next, the input molecular three-dimensional geometric structure is aligned through format conversion to obtain the target molecular descriptor; the generated molecular descriptor and the target molecular descriptor are used to calculate the loss through the loss function to further train the initial model and finally obtain the molecular generation model.

[0089] After the molecular generation model is fully trained, when using this method for molecular generation, the three-dimensional geometric structure of the organic molecule with the target material characteristics is first input into the cross-modal information extraction model, and then the obtained electronic structure cross-modal information and other set generation conditions are input into the molecular generation model to obtain new candidate molecules with electronic structure information similar to the organic molecule with the target material characteristics and meeting other generation conditions. The generated molecules are verified by structure verification and synthesizability verification, and then experimentally verified.

[0090] Preferably, on the basis of the above, the most representative part of the electronic structure of the generated candidate molecule can be further analyzed and judged based on expert knowledge, so as to optimize the frontier molecular orbitals contained in the cross-modal information and the fusion method of the cross-modal information. By adjusting the generation method and constraint method of the cross-modal information, the molecular generation model can be further optimized.

[0091] The following is a more specific application example for explanation.

[0092] A new material molecular design method based on the molecular orbital structure of TADF material comprises the following steps:

[0093] Step 1: Collect molecular structure data and form a data set, which contains common real organic molecules and TADF material molecules;

[0094] See also Figure 7 , Step 2: The first module of the cross-modal extraction model is obtained based on MPNN training, and the second module can be implemented using the self-attention mechanism model (Self-attention layers);

[0095] See also Figure 8 , Step 3: Sequence-based molecular generation model

[0096] This molecular generation model converts the cross-modal information of multiple molecular orbitals into dimensionally aligned sequence input units through the alignment model, and inputs them into the model in series. The basic structure prompts are used as other condition constraints, and are input into the model in series after the electronic structure condition constraints are input. After the generation of prompt words begins, the model generates the SMILES code of the target molecule.

[0097] See also Fig. 9 , Step 4: Material Design-Verification-Optimization Device

[0098] The generated TADF candidate materials are made into devices, and the experimental results are fed back to the device users with expert knowledge through material experimental verification devices. The users analyze the test results and adjust the constraints of the cross-modal information of the electronic structure of the molecular generation model. In this case, the number of molecular orbitals, the number of central atoms selected for each molecular orbital, and the global interaction characteristics of the central atoms {g i} j The improved molecular generation model is used to iteratively explore new TADF materials with target electronic structure properties.

[0099] The above describes the material design method based on the electronic structure in the embodiment of the present invention. The following describes the material design device based on the electronic structure in the embodiment of the present invention. Fig.10 , an embodiment of the material design device based on electronic structure in the embodiment of the present invention includes:

[0100] The first calculation module 10 is used to obtain a plurality of existing molecular structures, and calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures using a quantum chemical calculation method to obtain a molecular structure information data set;

[0101] A first construction module 20 is used to construct a cross-modal information extraction model based on the molecular structure information dataset based on the intramolecular group interaction relationship;

[0102] The second calculation module 30 is used to obtain the three-dimensional geometric structure of the target molecule, use the cross-modal information extraction model to calculate the cross-modal information of the target molecule according to the three-dimensional geometric structure of the target molecule, and integrate the obtained cross-modal information into the molecular structure information data set;

[0103] The second construction module 40 is used to construct a molecular generation model according to the molecular structure information data set, obtain the molecular information of the target material to be predicted, and use the molecular generation model to predict according to the molecular information of the target material to be predicted to obtain the target material molecule.

[0104] The above is a detailed description of the electronic structure-based material design device in the embodiment of the present invention from the perspective of modular functional entities. The following is a detailed description of the electronic structure-based material design device in the embodiment of the present invention from the perspective of hardware processing.

[0105] Fig.11A schematic diagram of the structure of a material design device based on an electronic structure provided in an embodiment of the present invention, the material design device 900 based on an electronic structure may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. Among them, the memory 920 and the storage medium 930 can be short-term storage or permanent storage. The program stored in the storage medium 930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the material design device 900 based on the electronic structure. Furthermore, the processor 910 can be configured to communicate with the storage medium 930, and execute a series of instruction operations in the storage medium 930 on the material design device 900 based on the electronic structure to implement the steps of the material design method based on the electronic structure provided in the above-mentioned method embodiments.

[0106] The electronic structure-based material design device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 960, and / or one or more operating systems 931, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Fig.11 The structure of the electronic structure-based material design device shown does not constitute a limitation of the electronic structure-based material design device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0107] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes the steps of the electronic structure-based material design method.

[0108] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the equipment or device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0109] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0110] It is understandable that those skilled in the art can make equivalent substitutions or changes based on the technical solution and inventive concept of the present invention, and all these changes or substitutions should fall within the protection scope of the claims attached to the present invention.

Claims

1. A material design method based on electronic structure, characterized in that: The steps include: Obtain multiple existing molecular structures, and use quantum chemical calculation methods to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set; Based on the interaction relationship between intramolecular groups, a cross-modal information extraction model is constructed according to the molecular structure information dataset; Obtain the three-dimensional geometric structure of the target molecule, use a cross-modal information extraction model to calculate the cross-modal information of the target molecule based on the three-dimensional geometric structure of the target molecule, and integrate the obtained cross-modal information into the molecular structure information dataset; A molecular generation model is constructed based on the molecular structure information data set to obtain the molecular information of the target material to be predicted, and the molecular generation model is used to perform prediction based on the molecular information of the target material to be predicted to obtain the target material molecules.

2. The electronic structure-based material design method according to claim 1, characterized in that: The method of obtaining a plurality of existing molecular structures and using a quantum chemical calculation method to calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures to obtain a molecular structure information data set specifically includes: The three-dimensional coordinate structure information of multiple organic molecules is obtained, and the charge distribution in the frontier molecular orbital of each organic molecule is calculated using quantum chemical calculation methods. A label is established for each atom of the organic molecule based on the calculation results to obtain a molecular structure information data set.

3. The electronic structure-based material design method according to claim 1, characterized in that: The cross-modal information extraction model is constructed based on the molecular structure information dataset based on the intramolecular group interaction relationship, specifically including: Setting a first module, the first module is used to extract the cumulative features of each atom in the three-dimensional geometric structure of the molecule; Setting a second module, the second module is used to calculate the electronic structure information of each atom according to the accumulated characteristics of each atom; Integrate the first module and the second module to obtain a basic model; The molecular structure information dataset was used to verify the prediction results of the basic model. The basic model was optimized according to the verification results to obtain a cross-modal information extraction model.

4. The electronic structure-based material design method according to claim 3, characterized in that: The first module is set up, and the first module is used to extract the cumulative features of each atom in the three-dimensional geometric structure of the molecule, specifically including: Setting a multi-level sampling structure to obtain element information and multi-level collective sampling results based on the three-dimensional geometric structure of the molecule; Setting a geometric structure feature extraction model for calculating multi-level atomic features based on element information and multi-level set sampling results; Sets the multi-level feature accumulation model used to calculate the cumulative features of each atom based on the multi-level atomic features.

5. The electronic structure-based material design method according to claim 3, characterized in that: The second module is set, and the second module is used to calculate the electronic structure information of each atom according to the accumulated characteristics of each atom, specifically including: Set up the inter-group attention interaction sub-model to calculate the global interaction features of each atom based on the accumulated features of each atom; Sets the electronic structure distribution output model used to calculate the electronic structure prediction results for each atom based on the global interaction characteristics of each atom.

6. The electronic structure-based material design method according to claim 5, characterized in that: The method of obtaining the three-dimensional geometric structure of the target molecule, using a cross-modal information extraction model, calculating the cross-modal information of the target molecule according to the three-dimensional geometric structure of the target molecule, and integrating the obtained cross-modal information into a molecular structure information dataset specifically includes: Generate multiple electron distribution prediction models for different frontier molecular orbitals based on the cross-modal information extraction model; Acquire the three-dimensional geometric structure of the target molecule, use a plurality of electron distribution prediction models for different frontier molecular orbitals, and make predictions based on the three-dimensional geometric structure of the target molecule to obtain a plurality of frontier molecular orbital charge distribution prediction values ​​for each atom; Sort multiple predicted values ​​of frontier molecular orbital charge distribution according to their numerical values, and use one or more atoms with the largest numerical values ​​as the central distribution atoms of the frontier molecular orbital of the target molecule; The global interaction features corresponding to the central distribution atoms are extracted, the predicted values ​​of the frontier molecular orbital charge distribution of the central distribution atoms are merged with the global interaction features to obtain cross-modal information, and the obtained cross-modal information is integrated into the molecular structure information dataset.

7. The electronic structure-based material design method according to claim 1, characterized in that: The method of constructing a molecular generation model according to the molecular structure information data set, obtaining molecular information of the target material to be predicted, and using the molecular generation model to perform prediction according to the molecular information of the target material to be predicted to obtain the target material molecule specifically includes: An initial model is constructed using a molecular structure information data set and molecular information of known materials, wherein the molecular information of the known materials includes a molecular skeleton, a group structure or a physicochemical property, and a predicted molecular descriptor is generated using the initial model; Convert the molecular three-dimensional geometric structure corresponding to the molecular structure information data set to obtain a target molecular descriptor, calculate the loss value between the target molecular descriptor and the predicted molecular descriptor, and use the loss value to train the initial model to obtain a molecular generation model; The molecular information of the target material to be predicted is obtained, and a molecular generation model is used to perform prediction based on the molecular information of the target material to be predicted to obtain the target material molecule.

8. A material design device based on electronic structure, characterized in that: include: The first calculation module is used to obtain a plurality of existing molecular structures, and calculate the electron distribution of the frontier molecular orbitals of the existing molecular structures by using a quantum chemical calculation method to obtain a molecular structure information data set; The first building module is used to build a cross-modal information extraction model based on the molecular structure information dataset based on the intramolecular group interaction relationship; The second calculation module obtains the three-dimensional geometric structure of the target molecule, uses a cross-modal information extraction model to calculate the cross-modal information of the target molecule according to the three-dimensional geometric structure of the target molecule, and integrates the obtained cross-modal information into the molecular structure information dataset; The second construction module is used to construct a molecular generation model based on the molecular structure information data set, obtain the molecular information of the target material to be predicted, and use the molecular generation model to predict based on the molecular information of the target material to be predicted to obtain the target material molecule.

9. A material design device based on electronic structure, characterized in that: comprising a memory and at least one processor, wherein the memory has computer-readable instructions stored therein; The at least one processor calls the computer-readable instructions in the memory to execute each step of the electronic structure-based material design method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, the steps of the electronic structure-based material design method according to any one of claims 1 to 7 are implemented.

Images