Neural Network Retrosynthesis for Efficient Reactant Set Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for retrosynthesis are inefficient and computationally costly, struggling with large search spaces and lacking consistency in predicting reactants for synthesizing target compounds.
Innovation Solution
A neural network system is employed to process candidate sets of reactants, using multiple subnetworks to generate probabilities and scores, enabling efficient and accurate determination of optimal reactants through a dual loss training mechanism, incorporating both forward and backward reaction directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional methods are used to search through all possible reactant combinations, then comprehensive coverage of candidate sets is achieved, but computational cost and time increase exponentially
Solution Approach 1:
The patent divides the retrosynthesis problem into two separate prediction tasks: forward reaction prediction (reactants → product) and backward reaction prediction (product → reactants). Each direction is handled by a dedicated neural network model, allowing independent optimization and training. This segmentation enables efficient searching by leveraging the structured predictions from both directions rather than exhaustively evaluating all possible reactant combinations.
2Measurement precision
If the number of reactants considered increases, then the accuracy of finding optimal reactant sets improves, but the search space grows exponentially
Solution Approach 1:
The patent implements a feedback mechanism where the forward prediction model generates expected products from reactants, which are then compared with the actual target compound. The backward prediction model uses the target compound to predict reactants, and these predictions are evaluated against the forward model's expectations. This bidirectional feedback loop guides the search toward high-probability reactant sets without exhaustive exploration, maintaining accuracy while controlling search space complexity.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing retrosynthesis using a neural network. One of the methods includes generating a prediction of a set of a plurality of predicted reactants that are combinable to generate a target compound, the generating comprising processing, for each of a plurality of candidate sets of reactants, a network input characterizing the candidate set using a neural network, determining, for each candidate set of the plurality of candidate sets, a score using the generated probabilities; and selecting a particular candidate set of one or more reactants using the determined scores.