Application of molecular mechanics polarization to fragment based drug design.

Polarization is a term that is often excluded from almost all virtual screening. Polarizability helps explain interactions between nonpolar atoms and electrically charged species. When studying fragments in FBDD these minor interactions could have large effect in changing how well a ligand will bind to its target. After including the polarizability terms in docking a validation set of ligands (Favia et al., 2011) with GLIDE, it improved the results the amount good docked poses (< 2 Å RMSD) by up to 12%. However some ligands were bound in incorrect poses. Further investigation was carried out with MD to observe if given enough time ligands bound in an incorrect pose would return to the binding site. In the first stages of investigating MD we ascertained if we could use GPUs to simulate larger systems and faster. After some performing some MD simulations in GROMACs we found that GPUs were an improved option and thus continued the simulation work with ACEMD which allowed multiple GPUs in tandem. After running the MD simulations for 200ns with atomic charges generated from the polarization the results we found were quite interesting. Some ligands would be trapped in their binding site but would fluctuate quite readily such as 2GVV. Some ligands showed that despite low RMSD they would be ejected from the binding site. In some cases the ligands would then attempt to return to their binding site. Ligands such as in 2CIX would show binding based on the breathing movement of the protein. Some ligands such as 1F5F or 1F8E bound tightly to their binding site during the MD, these ligands also enjoyed improved docking polarization with 0.1 – 1.0 Å improvement. These could be carried forward to become good candidates for experimental testing. Polarization is shown to have an overall positive effect improving binding data and if implemented with simple methods would have little opportunity cost to be added to modern FBDD methods.