5D-QSAR:  The Key for Simulating Induced Fit?

In this journal we recently reported the development and the validation of a four-dimensional (4D)-QSAR (quantitative structure-activity relationships) concept, allowing for multiple conformation, orientation, and protonation state representation of ligand molecules. While this approach significantly reduces the bias with selecting a bioactive conformer, orientation, or protonation state, it still requires a "sophisticated guess" about manifestation and magnitude of the associated local induced fit-the adaptation of the receptor binding pocket to the individual ligand topology. We have therefore extended our concept (software Quasar) by an additional degree of freedom--the fifth dimension--allowing for a multiple representation of the topology of the quasi-atomistic receptor surrogate. While this entity may be generated using up to six different induced-fit protocols, we demonstrate that the simulated evolution converges to a single model and that 5D-QSAR--due to the fact that model selection may vary throughout the entire simulation--yields less biased results than 4D-QSAR where only a single induced- fit model can be evaluated at a time. Using two bioregulators (the neurokinin-1 receptor and the aryl hydrocarbon receptor), we compare the results obtained with 4D- and 5D-QSAR. The NK-1 receptor system (represented by a total of 65 antagonist molecules) converges at a cross-validated r2 of 0.870 and a predictive r2 of 0.837; the corresponding values for the Ah receptor system (represented by a total of 131 dibenzodioxins, dibenzofurans, biphenyls, and polyaromatic hydrocarbons) are 0.838 and 0.832, respectively. The results indicate that the formal investment of additional computer time is well-returned both in quantitative and in qualitative values: less-biased boundary conditions, healthier (i.e., less inbred) model populations, and more accurate predictions of new compounds.