A new concept for multidimensional selection of ligand conformations (MultiSelect) and multidimensional scoring (MultiScore) of protein-ligand binding affinities

In this work, eight different scoring functions have been combined with the aim of improving I the prediction of protein-ligand binding conformations and affinities. The obtained scores were analyzed using multivariate statis tical methods to generate expressions,,with the ability (1) to select the b est candidate between different docked conformations of an inhibitor (Multi Select) and (2) to quantify the protein-ligand binding affinity (MultiScore ). By use of the docking program GOLD, 40 different inhibitors were docked into the active site of three matrix metalloproteinases (MMP's), yielding a total of 120 enzyme-inhibitor complexes. For each complex, a single confor mation of the inhibitor was selected using principal component analysis (PC A) for the scores obtained by the eight functions SCORE, LUDI, GRID, PMF_Sc ore, D_Score, G_Score, ChemScore, and F_Score. Binding affinities were esti mated based on partial least-squares projections onto latent structures (PL S) on the eight scores of each selected: inhibitor conformation. By use of this procedure, R-2 = 0.78 and Q(2) = 0.78 were obtained when comparing exp erimental and calculated binding affinities. MultiSelect was evaluated by a pplying the same method for selecting docked conformations for 18 different protein-ligand;complexes of known three-dimensional structure. In all case s, the selected ligand conformations were found to be very similar to the e xperimentally determined ligand conformations. A more general evaluation of MultiScore was performed using a set of 120 different protein-ligand compl exes for which both the three-dimensional structures and the binding affini ties were known. This approach: allowed an evaluation of MultiScore indepen dently of MultiSelect. The generality of the method was verified by obtaini ng Q(2) = 0.68 and Q(2) = 0.67, when comparing calculated and experimental binding affinities for the 120 X-ray structures. In all cases, LUDI, SCORE, GRID, and F_Score were included as important functions, whereas the fifth function was PMF_Score and ChemScore for the MMP and X-ray models, respecti vely.