Table 9. 5-Fold Cross-Validation Performance of Different Descriptors Using Gradient Boosted Trees and a RF Regressor on the BASF Data Set^a.

XGBoost
descriptor type	R2	MAE
MFP	0.716 ± 0.046	0.398 ± 0.018
GRADEOPD	0.448 ± 0.060	0.590 ± 0.025
MFP + GRADEOPD	0.704 ± 0.062	0.421 ± 0.045
GRADE	0.519 ± 0.061	0.551 ± 0.031
MFP + GRADE	0.716 ± 0.047	0.416 ± 0.023
RDKitPhysChem	0.659 ± 0.038	0.472 ± 0.027
MFP + RDKitPhysChem	0.716 ± 0.044	0.423 ± 0.028
PLEC	0.722 ± 0.018	0.402 ± 0.017
MFP + PLEC	0.732 ± 0.039	0.404 ± 0.035

RF Regressor
descriptor type	R2	MAE
MFP	0.665 ± 0.020	0.457 ± 0.020
GRADEOPD	0.483 ± 0.014	0.579 ± 0.016
MFP + GRADEOPD	0.668 ± 0.028	0.464 ± 0.026
GRADE	0.509 ± 0.022	0.565 ± 0.020
MFP + GRADE	0.661 ± 0.029	0.467 ± 0.027
RDKitPhysChem	0.587 ± 0.028	0.510 ± 0.027
MFP + RDKitPhysChem	0.643 ± 0.019	0.477 ± 0.025
PLEC	0.687 ± 0.020	0.448 ± 0.016
MFP + PLEC	0.689 ± 0.021	0.448 ± 0.016

^aOPD = Only Pose-Dependent. The best performing ones are marked by the use of bold letters.