Learning on compressed molecular representations

2024/11/04 by Jan Weinreich, Daniel Probst · 2 voices
Biochemistry, Genetics and Molecular Biology · #Fractal and DNA sequence analysis

paper · pdf · doi:10.1039/d4dd00162a

Abstract

It was proposed that a k -nearest neighbour classifier is able to outperform large-language models using compressed text as input and normalised compression distance as a metric. We successfully applied this method to cheminformatics tasks.

Citations

• SchNet – A deep learning architecture for molecules and materials

Discussions

• #compchem Learning on compressed molecular representations #compchemsky doi.org/10.1039/D4DD... [bsky, 18 points, 0 comments]
• 🥳Congratulations to Daniel Probst (@skepteis.bsky.social) & Jan Weinreich, winners of Digital Discovery’s Outstanding Early Career Researcher Award with their paper! 📖Read their full paper "Learnin [bsky, 13 points, 1 comments]

Related

• Molecular representation learning with language models and domain-relevant auxiliary tasks
• Constant Size Molecular Descriptors For Use With Machine Learning
• FragmentNet: Adaptive Graph Fragmentation for Graph-to-Sequence Molecular Representation Learning
• A Universal Non-Parametric Approach For Improved Molecular Sequence Analysis
• Large-Scale Chemical Language Representations Capture Molecular Structure and Properties