MolSight: Molecular Property Prediction with Images

TL;DR

MolSight demonstrates that simple 2D molecular images processed by vision models can effectively predict molecular properties, outperforming complex methods on multiple benchmarks with lower computational cost.

Contribution

This work is the first large-scale systematic study of vision-based molecular property prediction using 2D images and introduces a chemistry-informed curriculum for improved performance.

Findings

Achieves top results on 5 out of 10 benchmarks.

Single 2D images suffice for competitive property prediction.

Outperforms multi-modal methods with 80x lower FLOPs.

Abstract

Every molecule ever synthesised can be drawn as a 2D skeletal diagram, yet in modern property prediction this universally available representation has received less focus in favour of molecular graphs, 3D conformers, or billion-parameter language models, each imposing its own computational and data-engineering overhead. We present $\textbf{MolSight}$, the first systematic large-scale study of vision-based Molecular Property Prediction (MPP). Using 10 vision architectures, 7 pre-training strategies, and $2\,M$ molecule images, we evaluate performance across 10 downstream tasks spanning physical-property regression, drug-discovery classification, and quantum-chemistry prediction. To account for the wide variation in structural complexity across pre-training molecules, we further propose a $\textbf{chemistry-informed curriculum}$: five structural complexity descriptors partition the corpus into five tiers of increasing chemical difficulty, consistently outperforming non-curriculum baselines. We show that a single rendered bond-line image, processed by a vision encoder, is sufficient for competitive molecular property prediction, i.e. $\textit{chemical insight from sight alone}$. The best curriculum-trained configuration achieves the top result on $\textbf{5 of 10}$ benchmarks and top two on $\textbf{all 10}$, at $\textbf{$\textit{80$\times$ lower}$}$ FLOPs than the nearest multi-modal competitor.