Assessing drug-likeness--what are we missing?

The concept of drug-likeness helps to optimise pharmacokinetic and pharmaceutical properties, for example, solubility, chemical stability, bioavailability and distribution profile. A number of molecular descriptors have emerged as reasonably informative and predictive, for example, the Rule-of-Five. Here, we review some current approaches, then discuss their major shortcoming, namely the static nature of the structural features and physicochemical properties they encode. As we demonstrate, molecules are not 'frozen statues' but 'dancing ballerinas', and several of their computable physicochemical properties are conformation-dependent and lead to the concept of property spaces. Molecular sensitivity (namely, how much a given computable physicochemical property varies as a function of flexibility) appears as a promising descriptor to encode some of the information contained in molecular property spaces.