Self-Supervised Learning on Molecular Graphs: A Systematic Investigation of Masking Design
Jiannan Yang, Veronika Thost, Tengfei Ma
TL;DR
This work formalizes masking-based self-supervised learning for molecular graphs within a probabilistic framework, identifying three design dimensions: masking distribution, prediction target, and encoder architecture. Through rigorously controlled experiments and information-theoretic analyses, the authors show that sophisticated masking strategies offer little consistent benefit for common node-level tasks and come with substantial computational cost. In contrast, semantically richer prediction targets, especially motif-level supervision, yield substantial downstream gains when paired with expressive Graph Transformer encoders like GraphGPS. The findings argue for a shift in SSL design toward meaningful chemical semantics and encoder-target alignment, providing practical guidance for building more effective molecular representations.
Abstract
Self-supervised learning (SSL) plays a central role in molecular representation learning. Yet, many recent innovations in masking-based pretraining are introduced as heuristics and lack principled evaluation, obscuring which design choices are genuinely effective. This work cast the entire pretrain-finetune workflow into a unified probabilistic framework, enabling a transparent comparison and deeper understanding of masking strategies. Building on this formalism, we conduct a controlled study of three core design dimensions: masking distribution, prediction target, and encoder architecture, under rigorously controlled settings. We further employ information-theoretic measures to assess the informativeness of pretraining signals and connect them to empirically benchmarked downstream performance. Our findings reveal a surprising insight: sophisticated masking distributions offer no consistent benefit over uniform sampling for common node-level prediction tasks. Instead, the choice of prediction target and its synergy with the encoder architecture are far more critical. Specifically, shifting to semantically richer targets yields substantial downstream improvements, particularly when paired with expressive Graph Transformer encoders. These insights offer practical guidance for developing more effective SSL methods for molecular graphs.