MolHyper: Hypergraph-enhanced Graph Network for Accurate Molecular Property Prediction

Accurate prediction of molecular properties plays a crucial role in drug discovery, materials science, and elucidating chemical reaction mechanisms. Recently, the latest advances in deep learning have significantly outperformed traditional methods in this task. However, existing graph-based methods still face challenges in handling complex molecular structures, such as distinguishing molecules with similar structures but different functional groups. To address these issues, we propose a graph network with hypergraph properties, denoted as MolHyper, which mitigates the impact of noise and enhances prediction accuracy by capturing internode message passing and higher-order correlations of motifs. Experimental results on eight benchmark datasets from MoleculeNet demonstrate the effectiveness of this approach.