As an experimentalist, my bias is that it is not possible to validate a structure-activity relationship without an experimental measurement of activity. IC50 is one such measurement, and in many cases they are easy to produce, using an in vitro biochemical system such as an enzyme assay or a competitive ligand displacement assay. It is important to use a simplified, well-understood biochemical system to be sure that the activity measurement has a straightforward interpretation. For example, in an enzyme assay with a purified enzyme, it is usually possible to relate the IC50 directly to the dissociation constant of a non-covalent inhibitor, which is directly related to the free energy of ligand binding. Thus the computed free energy calculations can be validated by comparison to the IC50s of a set of compounds. (In contrast, the potency of compounds tested in more complex biological systems, such as cell lines, can be confused by other factors, such as permeability and efflux.)

Of course, it is necessary to have the compounds available for the experimental measurements, and to make sure that these compounds are of high purity and are the chemicals they are supposed to be. Furthermore, it is necessary, in the case of relatively weak inhibitors (micromolar range of IC50 and above), to ascertain that the mode of inhibition is what is being modeled by the QSAR. This requires further experimental measurements.

Thanks Adam B Shapiro