Skip to main content

Table 1 Candidate models obtained for log Pliver using the dataset reported by [11]

From: Visual analytics in cheminformatics: user-supervised descriptor selection for QSAR methods

Model

Predictive accuracy

Subset cardinality

#Frequent descriptors

#Descriptors shared with other model

M1 (ALOGP, Mor29u, AMW, Se, Pol)

R2 = 0.81

MAE = 0.15

RMSE = 0.20

5

2

2

M2 (ALOGP, SP15, RDF015v, RDF020e, H6v)

R2 = 0.76

MAE = 0.17

RMSE = 0.23

5

1

1

M3 (ALOGP, Mor29u, X4Av, ESpm15,Mor31e, Ui)

R2 = 0.79

MAE = 0.16

RMSE = 0.21

6

3

3

M4 (ALOGP, Mor29u, X4Av, DP06, QZZv, Mor02v, F01[C–C])

R2 = 0.79

MAE = 0.16

RMSE = 0.21

7

3

3

  1. The second column shows the predictive accuracy of the “best” model after applying 4-fold cross validation on three different methods (linear regression, decision trees, and neural networks). In this case, the best predictive accuracy for the four models was obtained by using a decision tree (M5P). The parameter setup and predictive accuracy for all methods are available in the Additional file 1: Table S1.