Table 2.

Comparison of enantioselectivity predictions between the SEMG-MIGNN model with other SOTA models

Data Splitting	DRFP	MFF	SEMG-MIGNN
Random 90/10	0.190 ± 0.010	0.183 ± 0.010	0.180 ± 0.010
Random 70/30	0.201 ± 0.010	0.212 ± 0.020	0.189 ± 0.010
Random 50/50	0.248 ± 0.030	0.227 ± 0.030	0.205 ± 0.020
Random 30/70	0.259 ± 0.030	0.243 ± 0.030	0.240 ± 0.020
Iminea	0.227 ± 0.005	0.226 ± 0.005	0.238 ± 0.005
Thiola	0.774 ± 0.020	0.726 ± 0.020	0.300 ± 0.010
Catalysta	0.565 ± 0.020	0.464 ± 0.020	0.294 ± 0.010
Transformationa	0.235 ± 0.005	0.264 ± 0.005	0.205 ± 0.005

Note: The best performance of each task is shown in bold. ^aThese data splitting tasks refer to the extrapolative predictions based on the scaffold splitting of the reaction components. Details are elaborated in Supplementary Fig. 22. RMSEs are in kcal mol⁻¹.