Hyper-Mol: Molecular Representation Learning via Fingerprint-Based Hypergraph

Table 4

The test ROC-AUC performance of different GNN backbones with atom-level and fingerprint-level structural information.

Tasks	Backbone	Atom-level	Fingerprint-level	Gain (%)
Binary	GCN	0.725	0.878	+21.1
	GIN	0.704	0.868	+23.3
Multilabel	GCN	0.623	0.759	+21.8
	GIN	0.631	0.753	+19.3