Skip to main content

Table 5 Training/testing results of a random forest model that uses features of mutations besides either a real or predicted wild-type protein-ligand binding affinity to predict the mutated protein-ligand binding affinity

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

Metric

real–real

real-pred

pred-real

pred–pred

R2 (5-fold CV)

0.90

0.90

0.89

0.89

MAE (5-fold CV)

0.35

0.35

0.38

0.38

MSE (5-fold CV)

0.25

0.25

0.28

0.28

RMSE (5-fold CV)

0.50

0.50

0.53

0.53

R2 (test set)

0.90

0.89

0.87

0.88

MAE (test set)

0.35

0.38

0.40

0.38

MSE (test set)

0.25

0.27

0.32

0.28

RMSE (test set)

0.50

0.52

0.56

0.52

RAE (test set)

0.28

0.30

0.32

0.30

RRSE (test set)

0.33

0.34

0.37

0.34

  1. The model was trained/tested in four scenrios. real–real: the model trained using real wild-type protein-ligand bindinf affinity data and tested on real binding affinities also. real-pred: the model is trained using real wild-type protein-ligand binding affinity data and tested on predicted binding affinities. pred-real: the model is trained using predicted wild-type protein-ligand binding affinity data and tested on real binding affinities. pred–pred: the model is trained using predicted wild-type protein-ligand binding affinity data and tested on predicited binding affinities