Multi-conformational Ligand Representation in 4D-QSAR: Reducing the Bias Associated with Ligand Alignment

Abstract

Quantitative structure-activity relationship (QSAR) is an area of computational research which builds mathematical or atomistic models to predict biological activities of molecules. While more powerful approaches make use of a genetic algorithm to reduce the bias with respect to model construction, the predictive power of the resulting surrogate still critically depends on the spatial alignment of the ligand molecules used to construct it. The 4D-QSAR concept Quasar developed at our laboratory not only takes local induced fit and H-bond flip-flop into account but also allows for the representation of the ligand molecules by an ensemble of conformations and/or orientations. The contribution of a single entity within this ensemble to the total ligand-receptor interaction energy is determined by a Boltzmann criterion. The three-dimensional surrogate is represented by a family of receptor-surface models, populated with atomistic properties—hydrogen bonds, salt bridges, hydrophobic particles, and solvent—mapped onto it. Quasar has been used to establish QSARs for the enzyme dopamine β-hydroxylase and for the aryl hydrocarbon receptor. The surrogates were able to predict free energies of ligand binding, ΔG°, for external sets of 15 and 26 test ligand molecules, respectively, to within 0.7 kcal/mol (rms) of the experimental value, with the largest individual deviation not exceeding 1.3 kcal/mol. The results indicate that the use of a multiple-ligand representation is superior to a single-conformer concept and reduces the user bias associated with the ligand alignment. Moreover, the selection protocol demonstrates that the technique is capable of identifying a small number of active conformations and does not prefer a larger selection of lesser-contributing entities.