Graph Convolutional Model for Catalyst and Solvent Efficacy Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current software solutions lack effective methods for predicting catalysts and solvents in chemical reactions, as existing datasets are incomplete and not machine-readable, hindering the development of models for catalyst and solvent prediction.
Innovation Solution
A computer-implemented system using a graph convolutional model with a multilayer fully connected neural network to predict catalysts and solvents by generating efficacy scores based on reaction data, including a scoring engine to calculate catalyst and solvent efficacy scores and a prediction engine to identify suitable catalysts and solvents for specific reactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing datasets are used for catalyst and solvent prediction, then the prediction model can be developed, but the datasets are incomplete and not machine-readable, leading to poor prediction accuracy
Solution Approach 1:
The patent transforms chemical reaction data from traditional textual/visual formats into structured machine-readable parameters including molecular graphs, reaction conditions, and efficacy scores. This parameter transformation enables complete and accurate representation of chemical data for machine learning models, directly resolving the data completeness issue while maintaining prediction accuracy.
Solution Approach 2:
The patent replaces traditional manual data collection and analysis methods with automated machine learning systems that process structured chemical data. The system uses graph convolutional networks and neural networks to automatically predict catalyst and solvent efficacy, substituting manual chemical expertise with computational algorithms that can process complete dataset information.
2Measurement precision
If a comprehensive reaction data model is created to improve prediction accuracy, then the model complexity increases, making the system harder to implement and compute
Solution Approach 1:
The patent divides the complex prediction task into separate models for catalyst prediction and solvent prediction. Each model processes specific types of input data (catalyst-specific features for catalyst prediction, solvent-specific features for solvent prediction), reducing the complexity of individual models while maintaining comprehensive prediction capabilities through modular architecture.
Solution Approach 2:
The patent introduces graph representation as an intermediary data structure that bridges chemical molecular information and machine learning processing. By converting molecular structures into graph formats with nodes and edges, the system creates a standardized intermediate representation that simplifies data processing while preserving complete chemical information for accurate predictions.
3Productivity
If machine-readable reaction data is generated to enable automated prediction, then prediction efficiency improves, but data processing and model training require significant computational resources
Solution Approach 1:
The patent performs preliminary data processing by pre-converting chemical reaction data into structured machine-readable formats including graph representations and efficacy scores before model training. This preliminary structuring of reaction data, catalyst information, and solvent information reduces the computational burden during actual prediction operations, as the data is already optimized for machine learning processing.
Data Source
AI summary
Embodiments described herein provide a computer-implemented method for predicting reaction constituents, comprising: generating a catalyst or solvent efficacy score based on reaction data; storing a graph convolutional model, generated based on the reaction data and the catalyst/solvent efficacy scores: wherein for a catalyst, graph representations of the catalyst and of a reactant and a product are inputs to the graph convolutional model; and similarly for a solvent; predicting the effectiveness of a specified catalyst/solvent based on a specified reaction definition and/or predicting a catalyst/solvent by determining whether the first and second reactions have an identical core transform.