Computational aqueous solubility prediction for drug-like compounds in congeneric series.

It was the aim of the present work to develop a quantitative structure-property relationship (QSPR) model for predicting the aqueous solubility of drug-like compounds in congeneric series. Lipophilicity combined with structural fragment information, fragmental based correction factors and congeneric series indices were used as descriptors for a principal component analysis (PCA) followed by multivariate partial least squares regression statistics (PLS). The derived PLS regression model for the prediction of solubility parameters was based on an in-house data set of 2473 drug-like compounds. The generated PLS model had a coefficient of determination (R(2))=0.844 and a root-mean-square (rms) error of 0.51 log units. It predicted the solubility of the test data set with a high degree of accuracy (R(2)=0.81). In addition, the PLS model was successful in predicting the solubility of new congeneric test sets when solubility values of corresponding scaffolds were accessible.