Structural Hierarchy and Geometry in Molecular Representation Learning
Abstract
Molecular self-supervised learning uses chemical structures to guide which molecular embeddings should be similar. We study whether explicitly encoding a molecule's Bemis-Murcko scaffold and using it to supervise the molecular embedding changes what the model learns. We further test whether this effect depends on the embedding geometry by comparing Euclidean and Lorentz contrastive objectives. Across two augmentation strengths, scaffold-supervised models consistently organize molecules according to both identical and structurally related scaffolds. The resulting embeddings also improve molecular property prediction on several tasks, while the exact gains depend on the predicted property. The effect of scaffold supervision on molecular organization is stronger under Lorentz objectives, but neither geometry provides a consistent overall advantage. These results show that explicitly teaching the relation between a molecule and its structural core can reliably shape the organization of molecular embedding space, while the extent of usefulness of this organization remains task dependent.
Related
- scDRP: disentangled representation learning for predicting single-cell responses to perturbations and estimating individual treatment effects
- Representation learning for low-data protein fitness prediction using sequence–structure embeddings and graph neural networks
- What to randomize matters: a controlled study of volume, axis and representation in low-data molecular generation
- Structural insights into an aptainhibitor of the SARS-CoV-2 E protein using molecular docking, molecular dynamics, and biophysical analyses
- STRUCTURAL AND MOLECULAR DYNAMICS ANALYSIS OF HUMAN β-DEFENSIN 1 VARIANTS REVEALS DISTINCT PERTURBATION SIGNATURES
BibTeX
@misc{sulu2026structural,
title = {{Structural Hierarchy and Geometry in Molecular Representation Learning}},
author = {David Sulu and Lorenzo Di Fruscia and Jana M. Weber},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2608.29886},
eprint = {2608.29886},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.29886},
}
RIS
TY - UNPB
TI - Structural Hierarchy and Geometry in Molecular Representation Learning
AU - David Sulu
AU - Lorenzo Di Fruscia
AU - Jana M. Weber
PY - 2026
DA - 2026/08/30
PB - arXiv
DO - 10.48550/arxiv.2608.29886
UR - https://arxiv.org/abs/2608.29886
AB - Molecular self-supervised learning uses chemical structures to guide which molecular embeddings should be similar. We study whether explicitly encoding a molecule's Bemis-Murcko scaffold and using it to supervise the molecular embedding changes what the model learns. We further test whether this effect depends on the embedding geometry by comparing Euclidean and Lorentz contrastive objectives. Across two augmentation strengths, scaffold-supervised models consistently organize molecules according to both identical and structurally related scaffolds. The resulting embeddings also improve molecular property prediction on several tasks, while the exact gains depend on the predicted property. The effect of scaffold supervision on molecular organization is stronger under Lorentz objectives, but neither geometry provides a consistent overall advantage. These results show that explicitly teaching the relation between a molecule and its structural core can reliably shape the organization of molecular embedding space, while the extent of usefulness of this organization remains task dependent.
ER -