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

Table 2

The test ROC-AUC performance of different models in binary classification benchmarks.

Datasets	HIV	BBBP	BACE	Avg.
GCN	0.740 ± 0.030	0.718 ± 0.009	0.716 ± 0.020	0.725
GIN	0.753 ± 0.019	0.658 ± 0.045	0.701 ± 0.054	0.704
N-Gram	0.830 ± 0.013	0.912 ± 0.030	0.876 ± 0.035	0.873
Hu et al.	0.802 ± 0.009	0.708 ± 0.015	0.859 ± 0.008	0.790
SchNet	0.702 ± 0.034	0.848 ± 0.022	0.766 ± 0.011	0.772
MGCN	0.738 ± 0.016	0.850 ± 0.064	0.734 ± 0.030	0.774
Hyper-Mol (GCN)	0.814 ± 0.011	0.922 ± 0.012	0.898 ± 0.009	0.878
Hyper-Mol (GIN)	0.808 ± 0.016	0.910 ± 0.022	0.885 ± 0.024	0.868

The numbers in bold represent the best performance.