3.2. Example: Solubility Data

Solubility is an important physical-chemical property in drug discovery. In the simplest definition, the solubility of a solute is the maximum amount of the solute that can dissolve in a certain amount of solvent. Intrinsic solubility is the solubility of the neutral form of the salt; the salt has not separated into ions. Some of the factors affecting solubility are: temperature, molecular weight, number of hydrogen bonds, polarity.

For a solvent to dissolve a solute, particles of the solvent must be able to separate the particles of the solute and occupy the intervening spaces. Water is a polar substance and polar solvents can generally dissolve solutes that are ionic. Dissolving takes place when the solvent is able to pull ions out of their crystal structure. Separation of ions by the action of a solvent is called dissociation; ions are dissociated by the water molecules and spread evenly throughout the solution.

Estimating solubility then becomes a bit more complicated when compounds have an ionized group because multiple forms of the compound exist with varying solubilities. The term aqueous solubility refers to the solubility of all forms of a compound. It is the sum of the solubility for the neutral compound plus the solubility of each ionized form of the compound; aqueous solubility is therefore, a function of pH. Since most drugs are either weakly acidic or basic, it is important to use the intrinsic solubility when building a computational model to predict solubility.

The example that we use in this chapter is a solubility data set (SOLUBILITY data set can be found on the book's companion web site). The solubilities were estimated using the old-fashioned method. An excessive amount of the compound was added to a flask of water; the flask was shaken and the amount of remaining compound measured. The amount of the compound dissolved is the solubility of that compound. In this data set there are 171 compounds. To limit the complexity of the example, only one software package was used to generate the molecular descriptors. The goal is to build an interpretable computational model that predicts the solubility of new compounds.