Chemical Structure Generation Using Graph Descriptors and Validity Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating chemical structures using neural networks are limited by their reliance on one-dimensional representations, which fail to fully express three-dimensional structures and thus neglect significant information, making it difficult to accurately predict properties and generate valid chemical structures.
Innovation Solution
A method and apparatus that utilize a chemical structure generation model, including an encoder and decoder, to transform descriptors into latent variables, with a validity evaluation model and property prediction model providing feedback for updating the generation model, ensuring the generation of valid chemical structures in a two-dimensional graph form.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If chemical structures are expressed as character strings or vectors, then the data processing is simple, but the three-dimensional structure information is lost and prediction accuracy decreases
Solution Approach 1:
The patent transforms chemical structure representation from one-dimensional character strings or vectors into two-dimensional graph structures. This dimensional change allows the model to preserve spatial relationships and structural information while maintaining computational tractability through graph neural network processing.
2Reliability
If reinforcement learning with validity evaluation is implemented, then the generation of valid chemical structures is improved, but the computational time and model complexity increase
Solution Approach 1:
The patent incorporates a validity evaluation model that assesses the chemical validity of generated structures during the training process. By performing preliminary validity checks and providing feedback rewards, the model learns to generate valid structures more efficiently, reducing the need for extensive post-generation validation and iterative corrections.
Data Source
AI summary
A method and apparatus for generating a new chemical structure using a chemical structure generation model are provided. The method includes receiving a first descriptor of a first chemical structure, encoding the first descriptor into a latent variable using an encoder of the chemical structure generation model, generating a second descriptor by decoding the latent variable using a decoder of the chemical structure generation model, generating the second chemical structure corresponding to the second descriptor.