Molecule Generation With Hard-Soft Fragment Retrieval Augmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional methods for generating molecules with desired properties are limited by the combinations of known molecule fragments, often failing to produce molecules with improved properties due to the inability to exceed the capabilities of existing fragments.

Innovation Solution

A computer-implemented method using a trained machine learning model to generate molecules by combining hard and soft molecule fragments, where hard fragments are included and soft fragments guide the generation process, enhanced by a molecular generative model that incorporates cross-attention and decoder layers to create novel molecules.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If known molecule fragments are combined to generate new molecules, then the generation process is systematic and database-searchable, but the generated molecules are limited to combinations of existing fragments and cannot exceed their capabilities

Engineering Contradiction:

Improvesystematic molecule generationVSAvoidmolecule property range

Core Design Contradiction:

Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The molecule is segmented into hard fragments (from database) and soft fragments (generated by AI model), allowing systematic retrieval of known fragments while enabling creative generation of novel fragments that extend beyond the database

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The molecule is constructed as a composite of hard fragments (providing structural foundation from known chemistry) and soft fragments (providing novel properties through AI-generated modifications), achieving both systematic generation and property enhancement

Inventive Principle:

Principle #40Composite materials

2Reliability

If trial and error experimentation is used to discover molecules, then all possible molecules can be tested, but the process is very time consuming and labor intensive

Engineering Contradiction:

Improvecomprehensive molecule testingVSAvoidexperimentation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The AI model performs preliminary generation of candidate molecules with desired properties before experimental testing, filtering and prioritizing promising candidates in silico to reduce the number of physical experiments needed

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system creates virtual copies of molecules through computational modeling and simulation, allowing extensive testing and evaluation of molecular properties in silico before physical synthesis and experimentation

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20250378919A1Techniques for generating molecules with fragment retrieval augmentation

Publication Date: 2025.12.11 NVIDIA CORP

AI summary

The disclosed method for generating molecules includes selecting, based on one or more molecule properties, one or more hard molecule fragments and one or more soft molecule fragments; and processing, using a trained machine learning model, the one or more hard molecule fragments and the one or more soft molecule fragments to generate a molecule, where the molecule includes the one or more hard molecule fragments, and the trained machine learning model generates the molecule based on the one or more soft molecule fragments.