Graph-Based Reaction Condition Prediction With Experimental Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for synthesizing chemical compounds require numerous experiments and substantial resources, making it difficult to identify stable synthetic conditions efficiently and achieve high yields.
Innovation Solution
A method utilizing a graph-type descriptor and a prediction neural network model to optimize synthetic conditions by determining combinations of reactants and target products, updating the model based on experimental results, and iteratively refining the conditions to achieve high yields.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If numerous experiments are performed to identify stable synthetic conditions, then the reliability of synthetic conditions improves, but the loss of time and resources increases
Solution Approach 1:
The patent applies preliminary action by training a neural network model beforehand with extensive experimental data to predict optimal synthetic conditions. This pre-computed knowledge base allows the system to recommend stable synthetic conditions without performing numerous new experiments, thus resolving the contradiction between reliability and time loss.
Solution Approach 2:
The patent implements feedback by iteratively updating the neural network model with experimental results. The model predicts synthetic conditions, experiments are performed, results are fed back to retrain the model, improving prediction accuracy over time. This reduces the number of experiments needed while maintaining high reliability of identified conditions.
2Manufacturing precision
If numerous experiments are performed to obtain high yield of target products, then the manufacturing precision improves, but the productivity decreases
Solution Approach 1:
The neural network model is trained in advance on large datasets of synthetic experiments to learn optimal conditions for high yield. This preliminary learning enables the system to predict conditions that achieve high manufacturing precision without requiring numerous new experiments, thus improving productivity.
Solution Approach 2:
The patent replaces the mechanical trial-and-error experimentation system with an intelligent prediction system based on neural networks. This substitution uses computational modeling to predict optimal synthetic conditions, achieving high yield precision while dramatically reducing the time and experimental iterations required, thereby improving productivity.
3Productivity
If a prediction neural network model is used to optimize synthetic conditions, then the productivity improves, but the measurement precision of predicted yields may be insufficient
Solution Approach 1:
The patent uses feedback by continuously updating the neural network model with actual experimental results. The model predicts yields, experiments are performed, actual results are fed back to retrain the model, improving measurement precision over time while maintaining high productivity through iterative refinement.
Solution Approach 2:
The model is preliminarily trained on extensive historical experimental data to establish accurate prediction capabilities before deployment. This pre-training ensures the model starts with reasonable measurement precision, which then improves further through feedback from new experiments, balancing productivity and accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of optimizing synthetic conditions includes receiving a graph-type descriptor comprising at least one of structural information of at least one reactant and structural information of a target product to be synthesized by the reactant; determining combinations of synthetic conditions for generating the target product by applying the graph-type descriptor to a prediction neural network model; selecting at least one initial condition combination from among the combinations based on a first confidence corresponding to a yield of the combinations; updating the prediction neural network model based on a ground-truth yield obtained from a result of an experiment with the initial condition combination; determining a priority of the combinations based on the updated prediction neural network model; and determining subsequent combinations of synthetic conditions based on the determined priority.