Search

KR-20260065055-A - PHARMACOPHORE-BASED 3D MOLECULE GENERATION SYSTEM FOR DRUG CANDIDATE DISCOVERY AND METHOD THEREOF

KR20260065055AKR 20260065055 AKR20260065055 AKR 20260065055AKR-20260065055-A

Abstract

The present invention relates to a method and system for generating a three-dimensional molecule based on a pharmacophore for deriving a new drug candidate, which efficiently designs and optimizes a three-dimensional ligand molecular structure capable of binding to a binding pocket of a target protein by considering protein-ligand interactions. The composition comprises the steps of collecting and inputting binding pocket data of a protein; recognizing and identifying the pharmacophore of a ligand; setting the pharmacophore as a fixed standard based on the binding pocket data and generating a three-dimensional structure of a ligand by applying an SE(3)-Equivalent Neural Network and a non-autoregressive diffusion generative model to form the remaining ligand structure; evaluating the binding affinity and structural similarity of the generated ligand; and optimizing the three-dimensional structure of the ligand according to the evaluation results to generate a final ligand.

Inventors

  • 박상현
  • 최승연
  • 서상민

Assignees

  • 연세대학교 산학협력단

Dates

Publication Date
20260508
Application Date
20241031

Claims (14)

  1. A method for generating a three-dimensional molecular structure of a ligand using a pharmacophore and diffusion generation model for deriving new drug candidates, A step of collecting and inputting protein binding pocket data; Step of recognizing and identifying the pharmacophores of the ligand; A step of setting pharmacophores as fixed standards based on the binding pocket data above, and generating a three-dimensional structure of the ligand by applying an SE(3)-Equivalent Neural Network and a non-autoregressive diffusion generative model to form the remaining ligand structure; A step of evaluating the binding affinity and structural similarity of the generated ligand; A method for generating a ligand molecular structure comprising the step of optimizing the three-dimensional structure of the ligand according to the above evaluation results to generate a final ligand.
  2. In paragraph 1, The above pharmacophor recognition and identification step is a method for generating a ligand molecular structure in which the ligand site with high interaction with the protein is recognized and identified using PLIP (Protein-Ligand Interaction Profiler).
  3. In paragraph 1, The above-described non-autoregressive diffusion generation model is a method for generating a ligand molecular structure that adds noise to the ligand through a forward process and gradually removes the noise through a reverse process to form a binding pocket and an optimized three-dimensional ligand structure.
  4. In paragraph 1, The above SE(3)-Equivalent Neural Network is designed so as not to be affected by the rotation and movement of the ligand, and is a method for generating a ligand molecular structure that forms a three-dimensional structure suitable for protein binding pockets.
  5. In paragraph 2, A method for generating a ligand molecular structure, wherein the above-mentioned pharmacophor recognition and identification step involves defining an index of ligand atoms important for binding as an interaction set using binding information collected from a protein-ligand complex, and selecting the position and binding state of atoms within the interaction set as pharmacophores.
  6. In paragraph 5, A method for generating a ligand molecule structure, wherein the above pharmacophor recognition and identification step involves decomposing a ligand molecule into a plurality of fragments, selecting a fragment with a high probability of binding to a protein using an atomic set (A i ) of each fragment and edge information (E i ) between atoms, and merging the fragments to form a pharmacophor.
  7. In paragraph 6, A method for generating a ligand molecular structure, wherein the above-mentioned pharmacophores recognition and identification step evaluates the interactions between fragments defined as pharmacophores to ultimately generate a list of pharmacophores that maximize binding affinity with a protein, and optimizes the remaining structure of the ligand using a non-self-regressive diffusion generation model based on the pharmacophores list.
  8. In a system for generating the three-dimensional molecular structure of a ligand using a pharmacophore and diffusion generation model for deriving new drug candidates, A data input section for inputting protein binding pocket data and pharmacopore data; Based on the above data, the pharmacophores of the ligand are recognized and identified, and A pharmacophor-based artificial intelligence model that sets the pharmacophor as a fixed standard based on the binding pocket data above, and generates the 3D structure of the ligand by applying an SE(3)-Equivalent Neural Network and a non-autoregressive diffusion generative model to form the remaining ligand structure; and A ligand molecular structure generation system comprising: a molecular structure output unit that outputs a three-dimensional structure of a ligand based on an evaluation of the binding affinity and structural similarity of the generated ligand.
  9. In paragraph 8, The above Pharmacopore-based artificial intelligence model is a ligand molecular structure generation system that recognizes and identifies ligand sites with high interaction with proteins using PLIP (Protein-Ligand Interaction Profiler).
  10. In paragraph 8, The above-described non-autoregressive diffusion generation model is a ligand molecular structure generation system that forms binding pockets and an optimized three-dimensional ligand structure by adding noise to the ligand through a forward process and gradually removing the noise through a reverse process.
  11. In paragraph 8, The above SE(3)-Equivalent Neural Network is a ligand molecular structure generation system designed so as not to be affected by the rotation and movement of the ligand, and forms a three-dimensional structure suitable for protein binding pockets.
  12. In Paragraph 9, The above-described pharmacophor-based artificial intelligence model is a ligand molecular structure generation system that uses binding information collected from a protein-ligand complex to define an indices of ligand atoms important for binding as an interaction set, and selects the positions and binding states of atoms within the interaction set as pharmacophores.
  13. In Paragraph 12, The above-described pharmacophor-based artificial intelligence model is a ligand molecule structure generation system that decomposes a ligand molecule into multiple fragments, selects a fragment with a high probability of binding to a protein using an atomic set (A i ) of each fragment and edge information (E i ) between atoms, and merges the fragments to form a pharmacophor.
  14. In Paragraph 13, A ligand molecular structure generation system in which the above-described pharmacorporate-based artificial intelligence model evaluates the interactions between fragments defined as pharmacorpors to ultimately generate a list of pharmacorpors that maximizes binding affinity with proteins, and optimizes the remaining structure of the ligand using a non-self-regressive diffusion generation model based on the said pharmacorporate list.

Description

Pharmacophore-Based 3D Molecular Generation System for Drug Candidate Discovery and Method Thereof The present invention relates to a method and system for generating three-dimensional molecules based on a pharmacophore for deriving new drug candidates, and more specifically, to a technology for efficiently designing and optimizing a three-dimensional ligand molecular structure capable of binding to the binding pocket of a target protein by considering protein-ligand interactions. The existing drug design and development process faced the problem of requiring significant time and resources, as well as incurring high costs, to search for and design candidate ligands capable of binding to target proteins. In particular, existing technologies primarily analyzed ligands by representing them as two-dimensional or one-dimensional structures, but this approach failed to adequately reflect protein-ligand interactions. Consequently, ligands with low binding affinity or instability were designed, which significantly reduced the efficiency of drug development. The "lock-and-key model," a traditionally used drug development model, predicted binding by relying on simple structural matchups between proteins and ligands. This approach described protein-ligand interactions solely from a geometric or topological perspective and failed to adequately reflect actual complex chemical interactions. In particular, it had limitations in that it did not consider specific pharmacophores of the ligands or account for non-covalent bonding and flexibility between the protein binding pocket and the ligand. Ligands generated by this approach were highly likely to fail to achieve predicted pharmacological efficacy in actual biological environments, and there was a significant risk of deriving new drug candidates with low binding affinity or inaccurate predictions. To address the aforementioned issues, recent attempts have been made to design ligand structures that reflect pharmacophores by developing 3D ligand generation models. Although 3D ligand models are designed to analyze protein-ligand interactions more precisely, existing models still suffer from problems such as failing to accurately recognize pharmacophores or adequately reflecting the complex interactions between pharmacophores and protein binding pockets. Furthermore, there are limitations in that the low efficiency of model training makes it difficult to utilize them in real-time during the new drug development process. Therefore, there is an increasing demand for technologies capable of accurately modeling complex chemical interactions between proteins and ligands. In particular, there is a need for technologies that can clearly recognize pharmacophores and efficiently design novel drug candidates capable of maximizing binding affinity with proteins based on them. To achieve this, it is necessary to develop new technologies that move beyond existing two-dimensional or one-dimensional approaches to precisely generate the three-dimensional structure of ligands and design ligand molecules optimized for specific protein binding sites. Figure 1 is a schematic diagram illustrating a pharmacopore-based three-dimensional molecule generation method and system for deriving new drug candidates. FIG. 2 is a diagram illustrating the configuration of a pharmacopore-based 3D molecule generation system according to an embodiment of the present invention. FIG. 3 is a diagram illustrating a method for generating and learning a pharmacophor-based artificial intelligence model according to an embodiment of the present invention. FIG. 4 is a diagram illustrating a method for a pharmacophor-based artificial intelligence model to generate a three-dimensional structure of a ligand according to an embodiment of the present invention. FIG. 5 is a diagram illustrating a 3D pharmacophores recognition and identification algorithm according to an embodiment of the present invention. FIG. 6 is a diagram illustrating the operation method of a pharmacophor-based artificial intelligence model according to an embodiment of the present invention. FIG. 7 is a diagram illustrating the visualization of a binding molecule by a pharmacophor-based artificial intelligence model according to an embodiment of the present invention. FIG. 8 is a diagram illustrating the process of generating a ligand molecule with a three-dimensional structure using a molecule generation system according to an embodiment of the present invention. FIG. 9 is a reference diagram evaluating the performance of a molecule generated by the PharDiff model (pharmacorporate-based artificial intelligence model) of a molecule generation system according to an embodiment of the present invention in terms of binding affinity and structural similarity. FIG. 10 illustrates a computing device that implements a descriptor generation method and a generation device according to an embodiment of the present invention. The present invention will be described