FraGAT

FraGAT, a Fragment-oriented Multi-scale Graph Attention Network, enhances molecular property prediction by incorporating the hierarchical structures of molecules, which existing graph-based models have largely overlooked. Recognizing the crucial role of functional groups within molecules in determining their physio-chemical properties and binding affinities, FraGAT represents molecules through graph fragments containing or constituting functional groups. This approach allows for a more nuanced and accurate prediction of molecular properties.

The development of FraGAT is underpinned by a novel definition of molecule graph fragments aimed at capturing the essential functional groups relevant to molecular properties.