Mol-SGCL: Molecular Substructure-Guided Contrastive Learning for Out-of-Distribution Generalization

Andrew Zhou, Yasha Ektefaie, Maha Farhat

Keywords: molecular property prediction, contrastive learning, substructure alignment, out-of-distribution generalization, drug discovery, small datasets

TL;DR: Mol-SGCL employs a triplet loss to align molecular representations with substructures that are plausibly causal for the target property, leading to improved OOD generalization on small datasets.

Submission Number: 40