SMPLIP-Score: predicting ligand binding affinity from simple and interpretable on-the-fly interaction fingerprint pattern descriptors

Table 1 The statistical performance of SMPLIP-RF models on PDBbind (Release 2015) according to different features compositions

Features	Train				Valid				Test
Features	RMSE	MAE	PCC	p_Value	RMSE	MAE	PCC	p_Value	RMSE	MAE	PCC	p_Value
IFP	0.537	0.420	0.977	0	1.372	1.066	0.724	4.17E-115	1.656	1.349	0.716	1.49E-29
IFP + Int-Dist	0.536	0.422	0.980	0	1.387	1.093	0.720	2.08E-112	1.692	1.388	0.711	5.15E-29
IFP + Frag	0.496	0.381	0.977	0	1.327	1.035	0.747	2.62E-125	1.489	1.227	0.771	8.71E-37
IFP + Int-Dist + Frag	0.494	0.382	0.978	0	1.346	1.054	0.740	8.07E-122	1.512	1.244	0.770	1.43E-36

The Refined set (n = 3481) used for training and validation, and core set (n = 180) as a test data. The boldface represents the model with better statistics from different features combination and max_features options
RMSE root-mean-square-error, MAE mean absolute error, PCC Pearson correlation coefficient, p_value p_value for statistical significance

ISSN: 1758-2946