Higher-Order Molecular Grammars for Generative and Foundation Models in Chemistry
Higher-order grammar sequences let molecular models encode ring topology with 100% validity and stronger MoleculeNet transfer.
The paper introduces Higher-order Grammar Representation (HGR), which lifts molecules to combinatorial complexes and serializes higher-order topology into compact context-free grammar rule sequences compatible with standard sequence models. To reduce bias toward simple rings, the authors build RingDiv, a 1.18-million-molecule ring-enriched benchmark including RingDiv300k, and define a ring diversity index. HGR generators achieve 100% validity by construction and rank first in Fréchet ChemNet Distance on all five generation benchmarks. HGR-FM records the highest mean AUC across seven MoleculeNet benchmarks, beating the strongest baseline by 8.3 and 3.3 AUC points under probing and full fine-tuning.
- HGR parses combinatorial complexes into context-free higher-order grammar rules.
- RingDiv holds 1.18 million molecules, with a curated RingDiv300k subset.
- HGR generation ranks first in FCD on all five benchmarks with 100% validity.
- HGR-FM leads mean AUC on seven MoleculeNet benchmarks under both transfer protocols.
- AI models
- HGR-FM
From the original article
1 min read at arxiv.orgMolecular learning models are strongly shaped by their underlying representations. Yet standard sequential and graph formalisms struggle to explicitly encode higher-order topology, such as ring systems and recurring motifs. Existing higher-order representations can capture these structures directly, but they are often computationally demanding and difficult to decode into valid molecules.
By Yiming Huang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.