The importance of the accuracy of the experimental data for the prediction of solubility

Aqueous solubility is an i mportant factor influencin g several a spects of the phar macokinetic profile of a drug. Nu merous p ublications pre sent di f- ferent methodologies for the de velopment of reliable computational models for the prediction of solubility fro m stru cture. The quality of s uch models c an be significantly affected by the accuracy of the em ployed experimental solubility data. In this work, the importance of the accuracy of the experimental solubility data used for mod el training was inv estigated. Three data sets were u sed as training sets - data set 1, containing solubility data collected from various lite- rature sources using a few criteria (n = 319), data set 2, created by substituting 28 valu es fro m data set 1 wit h unifor mly deter mined ex perimental d ata fro m one laboratory ( n = 319), and data set 3, cre ated by includi ng 56 ad ditional components, for which th e solubility was al so determined under uniform con- ditions in the same laboratory, in the d ata set 2 (n = 37 5). The sel ection of the most significant descriptors was performed by the heuristic method, using one- -parameter and multi-parameter analy sis. The correlations between th e most significant descriptors and solubility were established using multi-linear regres- sion analysis (MLR) for all t hree inve stigated data sets. Notable differen ces were observed between the equations corresponding to different data sets, sug- gesting that models updated with new experimental data need to be additionally optimized. It was succe ssfully shown that the inclusion of unifor m experimen- tal data co nsistently leads to an i mprovement in th e correla tion coefficie nts. These findings contribute to an emerging consensus that improving the reliabi- lity of solubility prediction requires the inclu sion of many diverse co mpounds for which solubility was measured under standardized conditions in the data set.