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

Table 3

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

Datasets	Tox21	SIDER	ClinTox	Avg.
GCN	0.709 ± 0.026	0.536 ± 0.032	0.625 ± 0.028	0.623
GIN	0.740 ± 0.008	0.573 ± 0.016	0.580 ± 0.044	0.631
N-Gram	0.769 ± 0.027	0.632 ± 0.005	0.855 ± 0.037	0.752
Hu et al.	0.787 ± 0.004	0.652 ± 0.009	0.789 ± 0.024	0.743
SchNet	0.772 ± 0.023	0.539 ± 0.037	0.715 ± 0.037	0.675
MGCN	0.707 ± 0.016	0.552 ± 0.018	0.634 ± 0.042	0.631
Hyper-Mol (GCN)	0.742 ± 0.038	0.659 ± 0.021	0.875 ± 0.078	0.759
Hyper-Mol (GIN)	0.723 ± 0.042	0.657 ± 0.026	0.879 ± 0.056	0.753

The numbers in bold represent the best performance.