Molecular Graph Contrastive Learning with Line Graph

2025-01-15

AI Summary

To address label scarcity in molecular property prediction and the semantic distortion and poor generalization inherent in existing graph contrastive learning (GCL) methods, this paper proposes LEMON: the first GCL framework for molecules that incorporates line-graph modeling to establish a synergistic contrastive paradigm between molecular graphs and their corresponding line graphs. It introduces an edge-attribute fusion module to preserve structural–chemical semantic consistency and a dual local contrastive loss to mitigate interference from hard negative samples. Crucially, LEMON eliminates reliance on handcrafted view augmentation. Extensive experiments on multiple standard molecular property prediction benchmarks demonstrate that LEMON consistently outperforms state-of-the-art view-generation methods, validating its effectiveness, robustness, and generalization capability—particularly under low-resource conditions.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Trapped by the label scarcity in molecular property prediction and drug design, graph contrastive learning (GCL) came forward. Leading contrastive learning works show two kinds of view generators, that is, random or learnable data corruption and domain knowledge incorporation. While effective, the two ways also lead to molecular semantics altering and limited generalization capability, respectively. To this end, we relate the extbf{L}in extbf{E} graph with extbf{MO}lecular graph co extbf{N}trastive learning and propose a novel method termed extit{LEMON}. Specifically, by contrasting the given graph with the corresponding line graph, the graph encoder can freely encode the molecular semantics without omission. Furthermore, we present a new patch with edge attribute fusion and two local contrastive losses enhance information transmission and tackle hard negative samples. Compared with state-of-the-art (SOTA) methods for view generation, superior performance on molecular property prediction suggests the effectiveness of our proposed framework.

Problem

• Graph Contrastive Learning
• Molecular Property Prediction
• Drug Design

Innovation

• LEMON method
• contrastive learning
• improved predictive performance