Title:Bi-level Contrastive Learning for Knowledge-Enhanced Molecule Representations

Abstract:Molecule representation learning is crucial for various downstream applications, such as understanding and predicting molecular properties and side effects. In this paper, we propose a novel method called GODE, which takes into account the two-level structure of individual molecules. We recognize that molecules have an intrinsic graph structure as well as being a node in a larger molecule knowledge graph. GODE integrates graph representations of individual molecules with multidomain biochemical data from knowledge graphs. By pre-training two graph neural networks (GNNs) on different graph structures, combined with contrastive learning, GODE fuses molecular structures with their corresponding knowledge graph substructures. This fusion results in a more robust and informative representation, which enhances molecular property prediction by harnessing both chemical and biological information. When fine-tuned across 11 chemical property tasks, our model outperforms existing benchmarks, registering an average ROC-AUC uplift of 13.8% for classification tasks and an average RMSE/MAE enhancement of 35.1% for regression tasks. Impressively, it surpasses the current leading model in molecule property predictions with average advancements of 2.1% in classification and 6.4% in regression tasks.