Application of descriptors based on Lipinski's rules in the QSPR study of aqueous solubilities

We complement new physically interpretable descriptors inspired by the Lipinski's rules of drug bioavailability with others obtained from the Dragon 3.0 software, in order to find the best QSPR relationship for aqueous solubilities of 100 structurally heterogeneous organic, drug-like compounds. The simultaneous linear regression analyses of 1367 variables lead to a six-parameter model containing two of the new proposed descriptors and which also possess good predictive ability given by R=0.8798 and cross-validated R(1-10%-o)=0.8199. We further validate the model found with an external test set composed of 48 compounds.