Molecular Graph Input Generation for Mixed-Component Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to efficiently predict the characteristics of multi-component substances with mixed components at various compounding ratios, as the three-dimensional structure is often unknown, making it difficult to use conventional techniques for accurate prediction.
Innovation Solution
An input data generation system that combines molecular graph data of multiple components with mixing rates to generate synthetic molecular graph data, converts it into a feature vector, and reflects mixing rates on the vector to create input data for machine learning, enabling accurate prediction of multi-component substance characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods using three-dimensional structure are used to predict substance characteristics, then prediction accuracy for single-component substances is improved, but the method becomes inapplicable to multi-component substances with unknown three-dimensional structures
Solution Approach 1:
The patent introduces molecular graph data as an intermediary representation that bridges the gap between known molecular structures and unknown multi-component substance structures. Instead of directly using three-dimensional structures, the system uses molecular graphs as a mediator that can be constructed from two-dimensional chemical structure data, enabling prediction for multi-component substances while maintaining prediction accuracy through graph-based machine learning models
2Adaptability or versatility
If molecular graph data is used as input for neural networks, then the method becomes applicable to multi-component substances, but the ability to efficiently handle various compounding ratios is not realized
Solution Approach 1:
The patent segments the molecular graph data into individual molecular components, each represented by its own molecular graph. This segmentation allows the system to independently process each component's structure and then combine them according to mixing ratios, enabling efficient prediction of multi-component substance characteristics by treating each molecule as a separate unit that can be recombined in various proportions
Solution Approach 2:
The patent changes the representation parameter from fixed three-dimensional structures to flexible molecular graphs that can accommodate variable mixing ratios. By representing molecules as graphs with nodes and edges that can be scaled and combined according to compounding ratios, the system efficiently handles different mixture compositions without requiring separate three-dimensional structure determination for each mixture
Data Source
AI summary
An input data generation system according to an embodiment includes at least one processor. The at least one processor receives at least an input of first molecular graph data specifying a molecular graph corresponding to a first molecule, second molecular graph data specifying a molecular graph corresponding to a second molecule, and mixing rate data indicating a mixing rate of each of the first molecule and the second molecule, generates synthetic molecular graph data by combining at least the first molecular graph data and the second molecular graph data, converts the synthetic molecular graph data into a feature vector, and generates input data for machine learning by reflecting the mixing rate data on the feature vector.