Local-Global Multimodal Contrastive Learning for Molecular Property Prediction

TL;DR

This paper introduces LGM-CL, a multimodal contrastive learning framework that combines molecular graphs and textual data to improve molecular property prediction, achieving strong results on MoleculeNet benchmarks.

Contribution

It presents a novel local-global multimodal contrastive learning approach that jointly models molecular structure and semantics for enhanced property prediction.

Findings

LGM-CL outperforms existing methods on MoleculeNet benchmarks.

The framework effectively integrates structural and semantic information.

Contrastive learning improves the robustness of molecular representations.

Abstract

Accurate molecular property prediction requires integrating complementary information from molecular structure and chemical semantics. In this work, we propose LGM-CL, a local-global multimodal contrastive learning framework that jointly models molecular graphs and textual representations derived from SMILES and chemistry-aware augmented texts. Local functional group information and global molecular topology are captured using AttentiveFP and Graph Transformer encoders, respectively, and aligned through self-supervised contrastive learning. In addition, chemically enriched textual descriptions are contrasted with original SMILES to incorporate physicochemical semantics in a task-agnostic manner. During fine-tuning, molecular fingerprints are further integrated via Dual Cross-attention multimodal fusion. Extensive experiments on MoleculeNet benchmarks demonstrate that LGM-CL achieves consistent and competitive performance across both classification and regression tasks, validating the effectiveness of unified local-global and multimodal representation learning.