Is the Data Set Suitable for Modeling

Solubility is a complex property, and this complexity confounds our ability to develop computational models to predict it. Most computational solubility models are empirical QSPR models, trained on solubility data sets either sourced from the literature and corporate databases or generated specifically for the purposes of modeling. Hence, it is not surprising that the quality of the computational model depends on the quality of the data set of experimental measurements used to train the model. 

The solubility measure describes the concentration reached in solution, when a pure phase of the material is allowed to dissolve in the solvent for a defined period of time, at a defined temperature (and pressure). Most often for pharmaceutical purposes, the pure phase is a solid, ideally a crystalline solid, and the liquid is water or a buffered aqueous solution, at a controlled temperature (often 25 or 37 °C) and ambient pressure. The purity of the solid can have a large effect on measured solubility. Solubility can be measured in water or in pH-controlled buffers. In water, the extent of solubility for ionizable compounds will depend upon the p fCa values and the nature of the counterion. In pH-controlled aqueous buffered solution, at equilibrium, the solubility will depend upon the compound s intrinsic solubility, its plQ, and the ionic strength. It may also depend upon the relative solubility of the initial added compound and the solubility of the salt formed by the compound with the buffer salts, with which the solid may equilibrate. In any buffer or solvent system, the measured solubility may depend on the time of sampling, as solubility kinetics 

---