Table 8.

Comparison of Q²_loo statistics of nD-QSAR methods for the property log K (CGB)^† for 31 (or 30)

nD-QSAR method	PCs/var.	Statistical method	${\text{Q}}^2$ loo	Equations/references
31/30 Steroids (all dataset)
Combined electrostatic and shape similarity matrix	6	Genetic NN	0.941	[59]
QuBiLS-MASc	6	MLR and GA	0.937	Equation 23
QuBiLS-MAS	6	MLR and GA	0.914	Equation 23
Hodking SM	6	Genetic NN	0.903	[59]
QuBiLS-MAS	5	MLR and GA	0.902	Equation 22
QuBiLS-MAS	4	MLR and GA	0.887	Equation 21
Fragment QS-SM	4	PLS	0.886	[60]
MEDV-13	5	MLR and GA	0.882	[61]
MiDSASA—“template”	2 “compounds”	–	0.88	[62]
SOMa	3	–	R2 0.85	[63]
Tuned-QSAR	6	MLR and PCA	0.842	[64]
Autocorrelation vector 30	–	–	0.84	[65]
CoMMA	3	PLS	0.828	[66]
QuBiLS-MAS	3	MLR and GA	0.826	Equation 20
Similarity Indices (ESP MC matrix 30)	1	PLS	0.820	[65]
SOMFA/esp + ALPHA	–	SOR	0.82	[67]
Combined electrostatic and shape similarity matrix	6	MLR and GA	0.819	[59]
EEVA	4	PLS	0.81	[68]
SOM-4D-QSAR	4	SOM neural network	0.80	[69]
Charges and Properties from MEPS-AM1	5	MLR	0.80	[70]
HE State/E-Statea,b	3	–	0.80	[71]
E-Statea,b	3	–	0.79	[71]
CoSA	3 “Bins”	PLS	0.78	[72]
QSAR/E-State	3 “atoms”	–	0.78	[73]
TQSI	4	MLR	0.775	[64]
EVA	5	PLS	0.77	[74]
CoMSA	1	PLS	0.76	[75]
MQSM	5	MLR and PCA	0.759	[64]
EVA + ALPHA	–	SOR	0.75	[67]
GRIND	–	PLS	0.75	[76]
SEAL	3	PLS	0.748	[77]
SOMFA/esp	6	PLS	0.74	[67]
CoSCoSAa	3	–	0.74	[78]
CoSASA	3 “atoms”	PLS	0.73	[72]
E-State and kappa shape index	4	MLR	0.72	[79]
TARIS	2	–	0.71	[80]
MQSM	3	MLR	0.705	[64]
Combined electrostatic and shape similarity matrix	5	PLS	0.70	[59]
SAMFA-RF	–	RF	0.69	[81]
SAMFA-PLS	4–5	PLS	0.69	[81]
4D-QSAR	2	PLS	0.69	[69]
CoMMA (ab initio)	6	PLS	0.689	[82]
QSARa	3	–	0.68	[83]
SOM-4D-QSAR	4	SOM Neural Network	0.68	[69]
Wagener’s (AMSP Method)	–	k-NN and FNN	0.630	[84]
SAMFA-SVM	–	SVM	0.60	[81]
ALPHA	2	PLS	0.57	[67]

Italic values indicate the results of QuBiLS-MAS approach

^aWhen it is applicable, specifies the number of components (PCs)

^b1.0 A models

^cCompound 31 excluded, taken as outlier, is not taken into account in the training set

^†Logarithm of the binding affinity to the corticosteroid-binding globulin (CBG)