Retrosynthetic Reagent Prediction via Neural Network Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques face difficulties in identifying and replacing reagents for synthesizing target compounds using retrosynthetic analysis, as they struggle to narrow down available candidates and determine suitable conversion reactions effectively.
Innovation Solution
An information processing program is developed that trains a model using vectors corresponding to target compounds and subcompounds, allowing for the calculation of similar subcompound vectors and reagent detection through a trained convolutional neural network or recurrent neural network, facilitating the identification of replaceable and cost-effective reagents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional retrosynthetic analysis techniques are used to identify reagents for target compound synthesis, then the synthetic pathway can be designed, but it becomes difficult to narrow down available reagent candidates and determine suitable conversion reactions effectively
Solution Approach 1:
The patent replaces manual retrosynthetic analysis with an automated computer-based system that uses machine learning models to predict reagents and reactions. The system automatically processes target compound structures, retrieves training data, trains prediction models, and identifies suitable reagents without requiring manual chemical expertise for each analysis case.
Solution Approach 2:
The system changes the approach from qualitative chemical reasoning to quantitative prediction by using molecular structures as input parameters for machine learning models. The models predict reagent properties and reaction outcomes based on numerical representations of chemical structures, enabling systematic evaluation of numerous reagent candidates.
2Ease of manufacture
If the number of available reagent candidates is increased to improve synthesis options, then more inexpensive and readily available reagents can be found, but narrowing down the candidates becomes more difficult
Solution Approach 1:
The system uses trained prediction models that have learned from historical synthesis data to evaluate and rank reagent candidates. The model provides feedback on which reagents are most likely to succeed based on their structural properties and predicted reaction outcomes, enabling precise selection from large candidate sets.
Solution Approach 2:
The system performs preliminary screening of reagent candidates using the trained prediction model before actual synthesis. This pre-evaluation filters out unsuitable reagents and prioritizes promising candidates, making the selection process more efficient and accurate when dealing with numerous available options.
Data Source
AI summary
A non-transitory computer-readable recording medium has stored therein an information processing program that causes a computer to execute a process including, executing training of a trained model based on training data defining relations between vectors corresponding to target compounds and vectors respectively corresponding to plural subcompounds included in synthetic pathways for manufacture of the target compounds and calculating vectors of plural subcompounds corresponding to a target compound to be analyzed by inputting a vector of the target compound to be analyzed into the trained model in a case where the target compound to be analyzed has been received.