Table 2. Comparison of Methods’ Performance for Different Training and Test Sets for Single Mutations Using Minimization Protocola.
method/energy function | training/test set | R(RCV) | RMSE (kcal mol–1) | slope |
---|---|---|---|---|
Pred1 | NM/NM | 0.63(0.61) | 1.22 | 1.00 |
NM/SKEMPI | 0.52 | 1.66 | 0.97 | |
SKEMPI/SKEMPI | 0.53(0.52) | 1.62 | 1.00 | |
SKEMPI/NM | 0.62 | 1.27 | 0.60 | |
Pred2 | NM/NM | 0.70(0.67) | 1.12 | 1.00 |
NM/SKEMPI | 0.57 | 1.61 | 0.95 | |
SKEMPI/SKEMPI | 0.58(0.58) | 1.55 | 1.00 | |
SKEMPI/NM | 0.69 | 1.20 | 0.97 | |
Pred4 | NM/NM | 0.74(0.68) | 1.05 | 1.00 |
NM/SKEMPI | 0.37 | 2.39 | 0.31 | |
CC/PBSA | NM/NM | 0.71 | 1.13 | 1.13 |
FoldX | test: NM | 0.47 | 1.60 | 0.53 |
test: SKEMPI | 0.37 | 2.14 | 0.42 | |
BeAtMuSiC | test: NM | 0.52 | 1.36 | 0.86 |
test: SKEMPI | 0.40 | 1.82 | 0.80 |
R is the Pearson correlation coefficient between experimental and predicted ΔΔG values and RCV is the five-fold cross-validated correlation coefficient. The last column shows the slope of the regression line between predicted and experimental values. CC/PBSA results were taken from the previous paper.22 All correlation coefficients are statistically significant (p-value ≪ 0.01).