A Novel Molecular Representation Learning for Molecular Property Prediction with a Multiple SMILES-Based Augmentation

Table 3

The AUROC and AUPRC scores of various approaches in HIV and BACE datasets. The predictive values of the approaches are partly derived from the related references [2, 8, 13, 32].

Model		HIV		BACE
		AUROC	AUPRC	AUROC	AUPRC
CheMixNet	CNN_RNN	0.8204	0.8614	0.7429	0.7162
SMILES-based	Smi2Vec-BiGRU	0.9117	0.8963	0.8440	0.7872
Conventional methods	XGBoost	0.7560	—	0.8500	—
	Multitask	0.6980	—	0.8240	—
Graph-based methods	GC	0.7630	—	0.7830	—
	Weave	0.7030	—	0.8060	—
	Pretraining GNN	0.7990	0.7806	0.8450	0.7908
3D-based models	Drug3D-Net	0.9621	0.9617	0.7185	0.6397
Our method (multiple SMILES)	RNN (one layer)	0.9567	0.9525	0.7879	0.7577
	RNN (two layers)	0.9613	0.9636	0.8083	0.7665
	CNN_RNN	0.9767	0.9798	0.8512	0.7919

The best results are highlighted in bold.