High-throughput, in silico prediction of aqueous solubility based on one- and two-dimensional descriptors.

An aqueous solubility model has been developed. The model is based solely on one- and two-dimensional descriptors and an artificial neural network to ensure fast execution. 63 descriptors expressing physicochemical and topological properties were used. The final model consisted of a training set of 3042 molecules, a test set of 309 molecules and an independent validation set of 307 molecules. The squared correlation coefficients were 0.91 for the training set, 0.89 for the test set and 0.86 for the independent validation set.