Augmentation-Free Molecular Structure Comparison for Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for computer-assisted assessment of chemical substances lack accuracy and efficiency in similarity and property predictions, particularly due to reliance on augmentations and insufficient consideration of cheminformatics and graphical structures.
Innovation Solution
A method that directly compares molecular structures to reference structures without augmentation, using latent representations and trainable parameters to achieve high similarity and property prediction accuracy through self-supervised learning (SSL) with loss functions like LSSL and cross-entropy loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If augmentation methods are used in self-supervised learning for molecular property prediction, then computational efficiency is improved, but prediction accuracy deteriorates
Solution Approach 1:
The patent extracts and removes the augmentation step from the self-supervised learning pipeline. Instead of using augmented molecular graphs (which improve computational efficiency but reduce accuracy), the invention directly compares original molecular structures to reference structures, eliminating the source of accuracy deterioration while maintaining the efficiency benefits of SSL.
Solution Approach 2:
The patent inverts the conventional SSL approach by not using data augmentation to create training variants. Instead, it uses the original molecular structure directly as the query and compares it to reference structures, reversing the typical augmentation-based paradigm while achieving both efficiency and accuracy.
2Device complexity
If conventional self-supervised learning methods are used without considering cheminformatics similarities, then training complexity is reduced, but prediction accuracy deteriorates
Solution Approach 1:
The patent applies local quality by considering cheminformatics similarities at the local molecular structure level. It compares specific molecular properties and structural features between query and reference molecules, allowing accurate predictions while maintaining manageable training complexity through focused local comparisons rather than global complex transformations.
Data Source
AI summary
An assessment method, a corresponding assessment program, a computer-readable data-carrier having stored thereon the assessment program, a computing device configured to carry out the assessment program and/or comprising the computer-readable data-carrier, and an assessment arrangement for a computer assisted assessment of at least one property parameter. Generally, the method includes obtaining a molecular structure of the chemical substance; obtaining at least one reference structure of at least one reference substance having at least one reference parameter, comparing the molecular structure to the at least one reference structure; and deriving the at least one property parameter from the comparison based on the at least one reference parameter.