A Genetic Algorithm Based Support Vector Machine Model for Blood-Brain Barrier Penetration Prediction

Table 2

Comparison of most relevant QSAR studies on BBB permeability.

Descriptors			Methods		Predictive accuracy on test set	Reference
Δlop , ,and	20	—	Linear Regression	0.69	—	Young et al. [77]
Excess molar refraction, dipolarity/polarisability, H-bond acidity, and basicity; Solute McGowan volume	148	30	LFER	0.75		Platts et al. [66]
Δ	55	—	Linear Regression	0.82	—	Lombardo et al. [78]
PSA, the octanol/water partition coefficient, and the conformational flexibility	56	7	MLR	0.85		Iyer et al. [79]
CODESSA/DRAGON (482)	200	110	PLS; SVM	0.83; 0.97		Golmohammadi et al. [62]
Molecular (CODESSA-PRO) descriptors ()	113	19	MLR	0.78		Katritzky et al. [15]
Molecular fragment (ISIDA) descriptors	112	19	MLR	0.90		Katritzky et al. [15]
PSA, , the number of H-bond acceptors, E-state, and VSA	144	10	Combinatorial QSAR (KNN; SVM)	0.91		Zhang et al. [17]
Abraham solute descriptors and indicators	328	—	LFER	0.75	—	Abraham et al. [51]
Abraham solute descriptors and indicators	164	164	LFER	0.71	, MAE = 0.20	Abraham et al. [51]
CODESSA/Marvin/indicator ()	260	63	GA based SVM	0.83	= 0.84, RMSE = 0.23	This research, GA/SVM, final model; = 13.3573, γ = 0.715761, ε = 0.151289
CODESSA/Marvin/indicator (236)	260	63	GA based SVM	0.97	= 0.55, RMSE = 0.31	This research, Grid/SVM; = 8.0, γ = 0.015625, ε = 0.0625
CODESSA/Marvin/indicator ()	260	63	GA based SVM	0.86	= 0.58, RMSE = 0.29	This research, Grid/SVM; = 8.0, γ = 1.0, ε = 0.125