Search

KR-102963337-B1 - METHOD AND APPARATUS FOR GENERATING CHEMICAL STRUCTURE

KR102963337B1KR 102963337 B1KR102963337 B1KR 102963337B1KR-102963337-B1

Abstract

A method and apparatus for generating a new chemical structure using a chemical structure generation model are provided. The method according to the present disclosure generates a second descriptor by encoding a first descriptor into a latent variable using an encoder and decoding the latent variable using a decoder, and can generate a new chemical structure corresponding to the second descriptor.

Inventors

  • 권영천
  • 강석호
  • 손원준
  • 이동선
  • 최윤석

Assignees

  • 삼성전자주식회사
  • 성균관대학교산학협력단

Dates

Publication Date
20260508
Application Date
20191119
Priority Date
20191010

Claims (18)

  1. In a method for generating a new chemical structure using a chemical structure generation model, The above method is performed by a processor, and A step of receiving a first descriptor for a predetermined chemical structure as input; A step of encoding the first descriptor into a latent variable using an encoder; A step of generating a second representation by decoding the above latent variable using a decoder; and A step of generating a new chemical structure corresponding to the second expressor above; Includes, The step of generating the above new chemical structure is, The step of inputting the second expressor into a validation evaluation model and receiving a reward as feedback from the validation evaluation model; and The method includes the step of updating the weight of the chemical structure generation model only when the compensation for the second descriptor is the first value, and The step of updating the weights of the chemical structure generation model above is, The step of inputting the second descriptor into a material property prediction model and receiving a predicted material property value as feedback from the material property prediction model; and A step of updating the weights of the chemical structure generation model based on the predicted physical property values; Includes, A method in which, in the above-described validity evaluation model, a first value is output as a reward when the second descriptor represents a valid chemical structure, and a second value is output as a reward when the second descriptor represents an invalid chemical structure.
  2. delete
  3. delete
  4. In Article 1, A method in which the first descriptor and the second descriptor have a two-dimensional graph form generated based on vertex information and edge information of a chemical structure.
  5. delete
  6. In Article 1, The step of receiving the above reward as feedback is, A step of determining whether the second expression includes a pre-set constraint (blacklist); and A step of inputting the second expressor that does not include the aforementioned preset constraints into the validation evaluation model, and receiving the reward as feedback from the validation evaluation model; A method including
  7. In Article 1, A method in which the above chemical structure generation model is a CVAE (Conditional Variational Autoencoder) model.
  8. In Article 1, The above validation model is a reinforcement learning model.
  9. In Article 1, A method in which the above-mentioned material property prediction model is a deep neural network model.
  10. In a method for generating a new chemical structure using a chemical structure generation model, The above method is performed by a processor, and A step of obtaining an arbitrary latent variable on the letant map of the chemical structure generation model above; A step of generating a representation by decoding the arbitrary latent variable using a decoder; and A step of generating a new chemical structure corresponding to the above-mentioned expression; Includes, The step of generating the above new chemical structure is, The step of inputting the above-mentioned expressor into a validation evaluation model and receiving a reward as feedback from the validation model; and The method includes the step of updating the weights of the chemical structure generation model only when the compensation for the above-mentioned descriptor is a first value, and The step of updating the weights of the chemical structure generation model above is, A step of inputting the above-mentioned descriptor into a material property prediction model and receiving a predicted material property value as feedback from the material property prediction model; and A step of updating the weights of the chemical structure generation model based on the predicted physical property values; Includes, A method in which, in the above-described validity evaluation model, a first value is output as a reward when the descriptor represents a valid chemical structure, and a second value is output as a reward when the descriptor represents an invalid chemical structure.
  11. In an apparatus for generating a new chemical structure using a chemical structure generation model, Memory in which at least one program is stored; and It includes a processor that executes at least one of the above programs, The above processor is, Using an encoder, a first descriptor for a given chemical structure is encoded as a latent variable, and A second representation is generated by decoding the above latent variable using a decoder, and It is to generate a new chemical structure corresponding to the second expression above, and The above processor is, Input the above second expressor into a validation evaluation model, and receive a reward as feedback from the above validation evaluation model, and The weights of the chemical structure generation model are updated only when the compensation for the second descriptor is the first value, and The above processor is, Input the above second descriptor into a material property prediction model, and receive a predicted material property value from the material property prediction model as feedback, The weights of the chemical structure generation model are updated based on the predicted physical property values above, and A device in which, in the above-described validity evaluation model, a first value is output as a compensation when the second descriptor represents a valid chemical structure, and a second value is output as a compensation when the second descriptor represents an invalid chemical structure.
  12. delete
  13. delete
  14. In Article 11, The device wherein the first descriptor and the second descriptor have a two-dimensional graph form generated based on vertex information and edge information of a chemical structure.
  15. delete
  16. In Article 11, The above processor is, Determine whether the above second expressor includes a pre-established constraint (blacklist), and A device that inputs the second expressor, which does not include the aforementioned preset constraints, into the validation evaluation model and receives the reward as feedback from the validation evaluation model.
  17. In an apparatus for generating a new chemical structure using a chemical structure generation model, Memory in which at least one program is stored; and It includes a processor that executes at least one of the above programs, The above processor is, Obtain an arbitrary latent variable on the letant map of the above chemical structure generation model, and A representation is generated by decoding the above arbitrary latent variable using a decoder, and It is to generate a new chemical structure corresponding to the above-mentioned expression, and The above processor is, Input the above-mentioned expressor into a validation model, and receive a reward as feedback from the above-mentioned validation model, and The weights of the chemical structure generation model are updated only when the compensation for the above-mentioned descriptor is the first value, and The above processor is, Input the above-mentioned descriptor into a material property prediction model, and receive predicted material property values as feedback from the material property prediction model, The weights of the chemical structure generation model are updated based on the predicted physical property values above, and A device in which, in the above-described validity evaluation model, a first value is output as a compensation when the descriptor represents a valid chemical structure, and a second value is output as a compensation when the descriptor represents an invalid chemical structure.
  18. A computer-readable recording medium having a program stored on it for executing the method of either claim 1 or claim 10 on a computer.

Description

Method and apparatus for generating chemical structure The present disclosure relates to an apparatus and method for generating a chemical structure. A neural network refers to a computational architecture that models the biological brain. As neural network technology advances, various types of electronic systems are utilizing neural networks to analyze input data and extract valuable information. Recently, research is actively underway to select chemical structures to be used in material development using neural network technology. Meanwhile, in the case of chemical structures expressed in the form of strings or vectors, there is a lot of information that is ignored and fails to fully represent the three-dimensional chemical structure. Accordingly, in order to overcome the limitations of string or vector forms and to predict physical properties more accurately and generate various chemical structures, a technology capable of utilizing three-dimensional chemical structures as input and output data is required. FIG. 1 is a block diagram illustrating the hardware configuration of a device according to one embodiment. FIG. 2 is a diagram illustrating operations performed in a deep neural network (hereinafter deep neural network) according to one embodiment. FIGS. 3a and 3b are drawings for explaining reinforcement learning and deep reinforcement learning (DRL) according to one embodiment. FIG. 4 is a diagram illustrating operations performed in a Conditional Variational AutoEncoder (CVAE) according to one embodiment. FIG. 5 is a drawing for explaining a two-dimensional graph-shaped representation according to one embodiment. FIG. 6 is a conceptual diagram illustrating the process of generating a chemical structure and predicting physical properties according to one embodiment. FIG. 7 is a diagram illustrating the process of training a chemical structure generation model according to one embodiment. FIG. 8 is a diagram illustrating the process of creating a new chemical structure using a chemical structure generation model according to one embodiment. FIG. 9 is a flowchart illustrating a method for generating a chemical structure in a chemical structure generation model according to one embodiment. Phrases such as "in some embodiments" or "in one embodiment" appearing in various places in this specification do not necessarily refer to the same embodiment. Some embodiments of the present disclosure may be represented by functional block configurations and various processing steps. Some or all of these functional blocks may be implemented by various numbers of hardware and/or software configurations that execute specific functions. For example, the functional blocks of the present disclosure may be implemented by one or more microprocessors or by circuit configurations for a specific function. Additionally, for example, the functional blocks of the present disclosure may be implemented in various programming or scripting languages. The functional blocks may be implemented as algorithms executed on one or more processors. Furthermore, the present disclosure may employ prior art for electronic configuration, signal processing, and/or data processing, etc. Terms such as “mechanism,” “element,” “means,” and “configuration” may be used broadly and are not limited to mechanical and physical configurations. Furthermore, the connecting lines or connecting members between the components depicted in the drawings are merely illustrative of functional connections and/or physical or circuit connections. In the actual device, connections between components may be represented by various alternative or added functional connections, physical connections, or circuit connections. The present disclosure will be described in detail below with reference to the attached drawings. FIG. 1 is a block diagram illustrating the hardware configuration of a device according to one embodiment. The device (100) can be implemented as various types of devices such as a PC (personal computer), server device, mobile device, and embedded device, and as specific examples, it may be a smartphone, tablet device, AR (Augmented Reality) device, IoT (Internet of Things) device, autonomous vehicle, robotics, medical device, etc. that perform voice recognition, image recognition, image classification, etc. using a neural network, but is not limited thereto. Furthermore, the device (100) may be a dedicated hardware accelerator (HW accelerator) installed in such devices, and the device (100) may be a hardware accelerator such as an NPU (neural processing unit), TPU (Tensor Processing Unit), or Neural Engine, which is a dedicated module for driving a neural network, but is not limited thereto. Referring to FIG. 1, the device (100) includes a processor (110) and a memory (120). Only the components related to the embodiments shown in FIG. 1 are shown in the device (100). Therefore, it is obvious to a person skilled in the art that the device (100) may include oth