. Author manuscript; available in PMC: 2011 Jul 26.

Published in final edited form as: J Chem Inf Model. 2010 Jul 26;50(7):1189–1204. doi: 10.1021/ci100176x

Table 3.

Statistical parameters of QSAR models obtained before and after curation.

ID	Name	R²	Q²	R²_EF	S_ws	S_cv	S_EF	R²_EVS	R²_EVS(NM)
1	Rat	0.96	0.84-0.93	0.89-0.92	0.11-0.13	0.16-0.24	0.20-0.26	–	–
2	Rat_(NM)	0.91-0.97	0.89-0.95	0.45-0.88	0.10-0.18	0.14-0.28	0.28-0.58	–	–
3	TP	0.83	–	0.76	0.33	–	0.38	0.54	−0.58
4	TP_(NM)	0.85	–	0.54	0.31	–	0.54	0.49	0.44
5	DILI non-curated	No modeling was possible
6	DILI⁵⁰	Modeling Set 5-fold external CV Accuracy = 62-68% External sets Accuracy = 56-73%
7^*	⁶²Ames non-curated	Sensitivity_RF=83%; Sensitivity_SVM=87%; Specificity_RF=Specificity_SVM=75% AUC_Gp=88%; AUC_SVM=89%; AUC_RF=83%
8^*	⁶³Ames curated	Sensitivity_RF=Sensitivity_SVM=79%; Specificity_RF=Specificity_SVM=81% AUC_GP=86%; AUC_SVM=84%; AUC_RF=83%

Where:

TP – Tetrahymena pyriformis dataset, _(NM) – modeling set with various representations of nitro groups

R² - determination coefficient, Q² - cross validation determination coefficient

R²_EF- determination coefficient for external folds extracted from the modeling set

S_ws - standard error of a prediction for work set

S_cv - standard error of prediction for work set in cross validation terms

S_ts - standard error of a prediction for external folds extracted from the modeling set

A - number of PLS latent variables, D - number of descriptors, M - number of molecules in the work set

R²_EVS - determination coefficient for external validation set

R²_EVS(NM) - determination coefficient for external validation set with shuffled nitro groups

AUC – Area Under Curve statistical parameter

RF – Random Forest

SVM – Supporting Vector Machine

GP – Gaussian Processes

Prediction performances are reported for external validation set.