PSnpBind-ML: predicting the effect of binding site mutations on protein-ligand binding affinity

Table 4 Training/testing results of a random forest model that uses features of proteins, ligands and binding sites to predict the wild-type protein-ligand binding affinity

Metric	BASR	PS	BSS	LS	LWSR	LVSR
R² (5-fold CV)	0.85	0.87	0.86	0.87	0.85	0.85
MAE (5-fold CV)	0.42	0.41	0.41	0.40	0.42	0.42
MSE (5-fold CV)	0.34	0.32	0.33	0.30	0.33	0.34
RMSE (5-fold CV)	0.58	0.56	0.57	0.55	0.58	0.58
R² (test set)	0.87	0.25	0.42	0.81	0.87	0.87
MAE (test set)	0.40	0.86	0.78	0.48	0.41	0.40
MSE (test set)	0.30	1.12	1.01	0.46	0.31	0.31
RMSE (test set)	0.55	1.06	1.00	0.68	0.56	0.56
RAE (test set)	0.33	0.85	0.75	0.39	0.33	0.33
RRSE (test set)	0.37	0.87	0.77	0.44	0.37	0.36

The model was trained/tested against six different data splits: BASR binding affinity-stratified random split, PS protein similarity-based split, BSS binding site similarity-based split, LS ligand similarity-based split, LWSR ligand weight-stratified random split, and LVSR ligand volume-stratified random split

ISSN: 1758-2946