Elucidating Molecular Overlays from Pairwise Alignments Using a Genetic Algorithm

Inferring the relative bioactive poses between active molecules is a common problem in drug discovery. The use of rapid pairwise alignment algorithms in conjunction with rigid conformer libraries has become a prevalent approach to this problem. These programs can be easily used to compare two molecules or suggest alternatives to a single known active. However, it is not obvious how to combine pairwise alignments between multiple actives into an overlay that reproduces the binding mode of those actives in the target receptor. We describe a new algorithm, DIFGAPE (DIstance geometry Focused Genetic Algorithm Pose Evaluator) that, given pairwise alignments of conformers of active compounds, attempts to reproduce overlays of ligand binding modes. The software was evaluated on 13 test systems from 9 protein targets using associated ligands extracted from the PDB. Starting from 2D ligand structures with no protein information, we were able in 4 systems to approximate the crystallographically observed binding mode. For example, the prediction for a set of 11 ligands targeting FXa had 1.6 A rmsd to crystal structure coordinates. Finally, the evaluation illustrated current challenges for molecular conformer generators and pairwise alignment algorithms.