Table 2.Statistical characteristics of QSAR models of NMDA receptor blockade
|
Model |
Descriptors |
r2 |
rmse |
r2cv |
rmsecv |
r2test |
rmsetest |
r2p |
|
MLR |
Σ(Ca)/α |
0.708 |
0.13 |
0.627 |
0.14 |
0.377 |
0.18 |
0.656 |
|
RF |
Σ(Ca), Eamax |
0.924 |
0.06 |
0.837 |
0.09 |
0.684 |
0.13 |
0.688 |
|
SVM |
Σ(Ca) |
0.990 |
0.02 |
0.858 |
0.09 |
0.751 |
0.12 |
0.699 |
|
GP |
α, Σ(Ca), Eamax |
0.971 |
0.04 |
0.748 |
0.12 |
0.815 |
0.10 |
0.747 |
Note. The following statistical methods were used to create the regression models: linear regression (LR), random forest (RF), support vector machine (SVM) realized in the corresponding computer programs: SVD [15], rf5new [16] and flssvm [17].