Neural Network Chemical Structure Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating chemical structures using neural networks are inefficient, error-prone, and computationally intensive, and often fail to accurately produce structures that satisfy multiple requirements.
Innovation Solution
A neural network device employing a deep neural network and a recurrent neural network, in conjunction with a conditional variational autoencoder, generates chemical structures by learning relationships between descriptors, properties, and structures, allowing for accurate, efficient, and resource-conserving production of chemical structures that meet specific target property and structure characteristic values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional methods are used to generate chemical structures, then the process can produce chemical structures, but the methods are inefficient, error-prone, and computationally intensive
Solution Approach 1:
The patent replaces traditional mechanical/computational methods for generating chemical structures with a neural network-based system. The neural network learns patterns from training data and generates chemical structures directly, substituting the previous computational approach that was inefficient and resource-intensive. This is evident in the abstract which describes using a neural network to 'generate chemical structures that satisfy various requirements' instead of traditional methods.
Solution Approach 2:
The neural network is trained on existing chemical structure data and then generates new structures by learning and copying patterns from the training data. The system copies the underlying patterns and relationships from the training set to generate novel chemical structures that satisfy target properties, rather than computing them from scratch using traditional methods.
2Manufacturing precision
If traditional neural network methods are used, then chemical structures can be generated, but they are incapable of accurately generating chemical structures that satisfy various metrics
Solution Approach 1:
The neural network performs preliminary learning during a training phase where it studies the relationships between chemical structures and their properties from training data. This preliminary action enables the network to accurately generate structures satisfying multiple metrics during actual use, as it has already learned the patterns and constraints beforehand.
Solution Approach 2:
The system uses feedback mechanisms where the neural network generates chemical structures and evaluates them against target properties and metrics. The training process incorporates feedback from the training data to improve the accuracy of structure generation, ensuring that generated structures satisfy multiple requirements reliably.
3Productivity
If conventional techniques are used for generating chemical structures, then the process can proceed, but it is incapable of generating chemical structures entirely in many cases
Solution Approach 1:
The patent replaces conventional computational techniques with a neural network system that has learned to generate chemical structures directly. This substitution dramatically improves the success rate of generation while reducing the time required, as the neural network can generate structures in a single pass rather than through iterative conventional methods that often fail.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of generating a chemical structure performed by a neural network device includes receiving a target property value and a target structure characteristic value; selecting first generation descriptors; generating second generation descriptors; determining, using a first neural network of the neural network device, property values of the second generation descriptors; determining, using a second neural network of the neural network device, structure characteristic values of the second generation descriptors; selecting, from the second generation descriptors, candidate descriptors that satisfy the target property value and the target structure characteristic value; and generating, using the second neural network of the neural network device, chemical structures for the selected candidate descriptors.