Neural Network Retrosynthesis Ranking for Reactant Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for synthesizing target products using neural networks face challenges in predicting optimal reactant combinations and determining experimental priority orders, leading to inefficiencies in chemical synthesis processes.

Innovation Solution

A method and apparatus utilizing a neural network to predict candidate reactant combinations and determine experimental priority orders through a pre-trained retrosynthesis prediction model, reaction prediction model, and yield rate prediction model, incorporating categorical latent variables to enhance the accuracy of chemical synthesis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing neural network methods are used to predict reactant combinations, then the synthesis process can be automated, but the accuracy in predicting optimal reactant combinations and determining experimental priority orders is insufficient

Engineering Contradiction:

Improveaccuracy of reactant combination predictionVSAvoidefficiency of synthesis process

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent divides the synthesis prediction task into multiple specialized neural network models: a retrosynthesis prediction model for generating candidate reactant combinations, a reaction prediction model for predicting products, and a yield rate prediction model for estimating reaction efficiency. This segmentation allows each model to specialize in a specific aspect, improving overall prediction accuracy while maintaining automated efficiency.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces categorical latent variables that represent different reaction types and conditions as additional parameters in the prediction process. By incorporating these latent variables into the neural network models, the system can capture nuanced chemical reaction characteristics, thereby improving the accuracy of reactant combination predictions and experimental priority ordering.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If multiple prediction models are used to improve accuracy, then the reliability of synthesis prediction increases, but the complexity of the system increases

Engineering Contradiction:

Improveaccuracy of synthesis predictionVSAvoidcomplexity of neural network system

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent integrates multiple specialized prediction models (retrosynthesis, reaction prediction, yield rate prediction) into a unified synthesis prediction system. The models work together in a coordinated manner, sharing the target product information and collectively generating ranked reactant combination recommendations. This merging approach maintains high prediction accuracy while organizing system complexity in a manageable and modular structure.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces categorical latent variables as intermediary elements that connect different prediction models. These latent variables serve as mediators that transmit reaction type and condition information between the retrosynthesis model, reaction prediction model, and yield rate prediction model, enabling coordinated operation without requiring direct complex interactions between all model components.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If traditional methods are used for determining experimental priority orders, then the process is simple, but the ability to prioritize reactant combinations based on multiple factors is limited

Engineering Contradiction:

Improveability to prioritize based on multiple factorsVSAvoidcomplexity of priority determination system

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent extends the priority determination system by incorporating multiple parameters including reaction yield rates, reaction types (represented by categorical latent variables), and predicted product match scores. The neural network models evaluate reactant combinations based on this multi-dimensional parameter set, enabling versatile priority ordering that considers various chemical and practical factors simultaneously.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional manual or rule-based priority determination methods with neural network-based prediction models. Instead of relying on simple heuristics or expert rules, the system uses trained neural networks to automatically evaluate and rank reactant combinations based on learned patterns from chemical data, providing more accurate and adaptable priority ordering.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS12547889B2Method and apparatus for synthesizing target products by using neural networks

Publication Date: 2026.02.10 SAMSUNG ELECTRONICS CO LTD

AI summary

An apparatus for generating a target product by using a neural network is configured to: predict candidate reactant combinations for generating the target product by using a pre-trained retrosynthesis prediction model; predict a prediction product with respect to each of the candidate reactant combinations by using a pre-trained reaction prediction model; and determine an experimental priority order of the candidate reactant combinations based on a result of comparing the target product with the prediction product.