Molecular Structure Inference Using Tree Decomposition and Site Information

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for generating molecular structures using generative models face challenges in efficiently learning and representing molecular structures, particularly when using tree representations, as they often result in irreversible conversions and increased computational costs due to the need for combined graph and tree representations.

Innovation Solution

An inferring device and training device that utilize tree decomposition with site information to enable reversible conversion between tree and graph representations, incorporating site information to maintain connection details and optimize the learning process, allowing for efficient generation and representation of molecular structures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If tree representation is used for molecular structure, then learning efficiency is improved, but reversibility of conversion is lost

Engineering Contradiction:

Improvelearning efficiencyVSAvoidreversibility of conversion

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent introduces site information as an intermediary element that mediates between the tree representation and the original molecular graph. This site information contains the necessary connection details that enable reversible conversion while maintaining the simplified tree structure for efficient learning.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the molecular structure representation into two parts: the tree structure for efficient learning and the site information for reversibility. This segmentation allows each component to serve its specific function without compromising the other.

Inventive Principle:

Principle #1Segmentation

2Reliability

If combined graph and tree representation is used for reversible conversion, then reversibility is improved, but computational cost increases

Engineering Contradiction:

Improvereversibility of conversionVSAvoidcomputational cost

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent extracts only the essential connection information needed for reversibility and stores it as site information within the tree structure. This extraction approach avoids the need to maintain the full graph representation, significantly reducing computational cost while preserving reversibility.

Inventive Principle:

Principle #2Taking out (Extraction)

3Reliability

If graph representation is used for molecular structure, then reversibility is maintained, but learning efficiency decreases

Engineering Contradiction:

Improvereversibility of conversionVSAvoidlearning efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

Instead of converting from graph to tree and losing information, the patent inverts the approach by building the tree structure first and then adding site information to enable reversibility. This inversion allows the benefits of tree representation to be realized while maintaining the ability to reconstruct the original graph.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentUS20230196075A1Inferring device, training device and inferring method

Publication Date: 2023.06.22 PREFERRED NETWORKS INC

AI summary

An inferring device includes one or more memories and one or more processors. The one or more processors are configured to generate information on a tree including information on a node and information on an edge from a latent representation by using a trained inference model; and generate a graph from the information on the tree. The information on the tree includes connection information on the nodes.