Table 3 Mean Absolute Error (MAE) and Geometric Mean Fold Error (GMFE) calculated for each combination of input descriptor set and the best regression model for that descriptor set, when predicting log Vss for all compounds in the external set (with 202 compounds)
Model | Input descriptor set | Regression model type | MAE | GMFE |
|---|---|---|---|---|
1 | predicted log Kt:p’s and molecular descriptors | M5P – regression tree | 0.4172 | 2.61 |
2 | molecular descriptors only (i.e., without predicted log Kt:p’s) | Bagging (a set of M5P model trees) | 0.3676 | 2.33 |
3 | descriptor set selected by genetic search-based CFS | Bagging (a set of M5P model trees) | 0.3609 | 2.29 |