Table 2.

Molecular property prediction performance on regression benchmarks.

Datasets	ESOL	FreeSolv	Lipophilicity	QM7	QM8	QM9
Metrics	RMSE	RMSE	RMSE	MAE	MAE	MAE
GraphSAGE	2.575	5.051	1.212	164.062	0.0388	11.178
GPT_GNN	1.612	5.284	0.820	229.053	0.0204	7.976
AttributeMask	1.439	8.062	0.784	261.588	0.0188	13.461
ContextPred	1.430	8.616	0.838	243.551	0.0205	16.886
InfoGraph	1.380	31.118	0.926	292.601	0.0192	12.350
MoCL	1.425	3.233	0.998	198.215	0.0903	NA
GraphLoG	1.390	4.515	0.857	274.071	0.0193	11.484
GraphCL	1.265	5.569	0.782	285.967	0.0199	9.773
JOAO	1.355	4.280	0.771	270.839	0.0206	22.507
MolCLR	1.333	3.285	0.720	104.184	0.0187	23.226
G_Motif	1.286	4.432	0.779	222.957	0.0203	11.065
MGSSL	1.346	2.980	0.751	155.913	0.0198	21.538
HiMol(SMALL)a	0.938	3.215	0.709	96.776	0.0196	3.770
HiMol(LARGE)b	0.833	2.283	0.708	91.501	0.0199	3.243

^aSMALL version implements 3-layer GIN as the GNN backbone.

^bLARGE version implements 5-layer GIN as the GNN backbone.

The values in bold highlight the best performing results of each benchmark.