Table 1 Reported performances of the thermodynamic solubility models published from 1997 to 2007.

From: Will we ever be able to accurately predict solubility?

Year	Reference	Descriptors	Size	Dataset	Method	RMSE	R2
1997	Huuskonen et al.⁸⁷	Electrotopological / Topological	83	Litterature	ANN	—	0.84
2000	Huuskonen et al.¹⁶	Structural	694	Khune et al.	MLR	—	0.67
					MLR		0.87
					ANN		0.85
					ANN		0.84
2001	Tetko et al.²³	Molconn-Z	1,291	Huuskonen et al.	MLR	0.81	0.85
	Tetko et al.²³	Molconn-Z	1,291	Huuskonen et al.	ANN	0.66	0.9
	Ran et al.⁴²	Melting Point / cLogP	380	AquaSol	GSE	0.76	—
	Bruneau⁸⁸	2D/3D/Charge/ Katrizky	2,233	Huuskonen et al.	ANN	0.82	—
	Liu et al.⁸⁹	2D Topological	1,312	Huuskonen et al.	ANN	0.71	—
2002	Klamt et al.⁹⁰	QM	257	QikProp dataset	MLR	0.61	—
	Engkvist et al.⁹¹	1D/2D Descriptors	1,290	Huuskonen et al.	ANN	—	0.95
	Chen et al.⁹²	Dipole, PSA, Vol, MW, Rot. & H-acc/don and D	321	Litterature	MLR	0.86	0.71
2003	Wegner & Zell⁹³	2D Topological	1,290	Huuskonen et al.	ANN	0.54	—
	Cheng & Merz⁹⁴	Cerius	2,440	AquaSol, PhysProp, Merck Index, PDR, CMC	MLR-GA	1.01	—
	Yan & Gasteiger⁹⁵	PETRA	1,293	Huuskonen et al.	MLR	—	0.89
	Yan & Gasteiger⁹⁵	PETRA	1,293	Huuskonen et al.	ANN	—	0.94
	Lind & Maltseva⁹⁶	Electrostatic, QM & topological	1,296	Huuskonen et al.	SVM	0.68	0.89
2004	Yan et al.⁹⁷	PETRA	2,084	Huuskonen et al.	ANN	—	0.94
	Hou et al.⁹⁸	2D Topological	1,299	Huuskonen et al.	MLR	—	0.9
	Fröhlich et al.⁹⁹	MOE & JOElib	1,297	Huuskonen et al.	SVM	—	0.9
	Votano et al.¹⁰⁰	Fragments & Counts	4,115	Aquasol, Physprop, PDR, Taskinen, Tetko, Lobell	MLR & PLS	—	0.84
			4,115		ANN		0.84
			1,840		ANN		0.86
	John S. Delaney¹⁹	cLogP, MW & Count	2,874	Abraham, Pesticide Manual, Syngenta	ESOL	—	0.55
2005	Matthew Clark¹⁰¹	2D descriptors	3,724	PhysProp	PLS	—	0.84
	Catana et al.¹⁰²	MOE, E-state & ISIS key	1,107	Pfizer proprietary & Public	PLS	0.48	0.94
					Non-Linear PLS
					NN
2006	Hansen et al.⁴³	MOE 2D/3D	4,569	PhysProp	ANN	0.97	0.94
	Wassvik et al.¹⁰³	Tm, LogP, Sm, Hm & Molconn-Z	428	Astrazeneca	GSE	0.92	0.73
	Wassvik et al.¹⁰³	Tm, LogP, Sm, Hm & Molconn-Z	428	Astrazeneca	Mod. GSE	0.73	0.78
2007	Wang et al.¹⁰	3D Topological, cLogP, MW & Count	1,878	Delaney et al., Huuskonen et al., Hou et al.	MLR	0.74	0.9
	Johnson et al.⁴⁵	VOLSURF	362	Literature	MLR & HH	0.61	0.88
	Schwaighofer et al.¹⁰⁴	Dragon	1,290	Huuskonen et al.	GP	0.55	0.93
	Schwaighofer et al.¹⁰⁴	Dragon	4,597	Huuskonen et al. & Others	GP	0.55	0.91

ANN: Artificial Neural Network
ASE: Abraham Solvation Equation
CNN: Convolutional Neural Network
CPANN: Count-Propagation Artificial Neural Network
DNN: Deep Neural Network
D-GIN: Directed GIN
D-MPNN: Directed-MPNN
GIN: Graph Isomorphism Network
GP: Gaussian Process
GNN: Graph Neural Network
GSE: General Solubility Equation
HH: Henderson-Hasselbalch equation
KNN: Kernel Neural Network
LS-SVM: Least-Square Support Vector Machine
MAT: Molecule Attention Transformer
MK: Multi Kernel
MLR: Multi Linear Regression
MLR-GA: Multi Linear Regression Genetic Algorithm
MPNN: Message Passing Neural Network
NFP: Neural FingerPrint
NL-PLS: Non-Linear Partial Least Square
PLS: Partial Least Square
RF: Random Forest
RM: Replacement Method
SMILES: Simplified Molecule Input Line Entry System
SNN: Shallow Neural Network
SR: Stepwise regression
SVM: Support Vector Machine
SVR: Support Vector Regression
TE: Theorical Equation
UG-RNN: Undirected Graph Recurrent Neural Network
CR: Contracted Ring
LMO: Leave-Many-Out
LOO: Leave-One-Out

Back to article page

Table 1 Reported performances of the thermodynamic solubility models published from 1997 to 2007.

Search

Quick links