Neural Network Synthesis Recipe Retrieval for Missing Reaction Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for discovering new materials and drugs are time-consuming and costly, and AI algorithms struggle to find novel synthesis recipes due to reliance on researcher intuition and biased experimental data, often missing critical information.
Innovation Solution
A neural network model is trained to generate synthesis recipes by extracting candidate embedding vectors based on similarity scores, using graph neural networks and contrastive learning to handle missing data and user feedback, enabling efficient prediction of chemical reactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional experimental methods are used for material discovery, then comprehensive data can be obtained through researcher intuition, but the process is time-consuming and costly
Solution Approach 1:
The neural network model is pre-trained on existing synthesis recipe data before actual use, enabling it to quickly predict outcomes for new queries without requiring time-consuming traditional experimentation. The model performs preliminary learning from historical data, allowing rapid prediction phase.
Solution Approach 2:
The system creates virtual representations (embedding vectors) of chemical compounds and reactions that mirror real-world properties. These digital twins allow simulation and prediction of synthesis outcomes without physical experimentation, dramatically reducing time and cost while maintaining reliability through accurate modeling.
2Productivity
If AI algorithms are used to search for synthesis recipes, then the process is faster and more automated, but the algorithms cannot easily find new synthesis methods for novel molecules
Solution Approach 1:
The embedding vectors capture local structural features and properties of individual molecules, allowing the neural network to recognize subtle patterns that indicate novel synthesis pathways. Each molecule's local chemical characteristics are preserved in its embedding representation, enabling discovery of new methods for novel molecules.
Solution Approach 2:
The system transforms chemical structures into high-dimensional embedding vectors, adding dimensional depth to the representation. This dimensional transformation allows the neural network to identify complex patterns and relationships that are not apparent in traditional 2D structural representations, enhancing novelty discovery capability.
3Ease of operation
If researcher knowledge is used to fill missing experimental information, then experiments can be conducted, but the results depend on background knowledge and lack objectivity
Solution Approach 1:
The neural network model automatically fills in missing experimental information by predicting it from patterns learned in the training data, eliminating the need for researcher intervention. The system serves itself by autonomously completing incomplete synthesis recipes with predicted values, ensuring objectivity through consistent algorithmic processing.
Solution Approach 2:
The model uses feedback from training data to continuously improve its predictions of missing information. By learning from historical experimental outcomes, the system refines its ability to objectively predict missing parameters, reducing dependency on subjective researcher knowledge while maintaining experiment conductability.
Data Source
AI summary
A method performed by at least one processor, includes receiving a query comprising at least one of a product, a reactant, a reagent, and a reaction condition; generating a query embedding vector corresponding to the query by inputting the query into a neural network model; extracting a candidate embedding vector from among a plurality of embedding vectors based on a similarity between the query embedding vector and the plurality of embedding vectors corresponding to reaction records stored in a database; and outputting a synthesis recipe corresponding to the query by retrieving a candidate reaction record corresponding to the candidate embedding vector in the database.