With the increasing power of computers, computational studies have become more and more significant in drug discovery. High binding free energy is one of the major requirements for an effective drug molecule, hence much effort has been spent to develop fast and accurate computational free energy methods. In this thesis, different free energy methods, i.e. umbrella sampling, thermodynamic integration, and double decoupling method, are applied to different systems related to drug discovery. For the first study, umbrella sampling studies are performed to calculate the absolute binding free energies of host-guest systems which serve as great model systems to assess free energy methods due to the small size of the systems, etc. We find that benchmarking our method with known systems can significantly improve the results for the unknown systems: the overall RMSE of the binding free energy versus experiment is reduced from 5.59 kcal/mol to 2.36 kcal/mol. The source of error could be from the un-optimized force constants used in umbrella sampling (hence possibly poor window overlaps), as well as force field, sampling issues, etc. Our results ranked 4th best in the Statistical Assessment of the Modeling of Proteins and Ligands (SAMPL6) blind challenge. For the second study, GPU accelerated thermodynamic integration (GPU-TI) is used to compute the relative binding free energies of a protein-ligand dataset originally assembled by Schrodinger, Inc. The calculations of relative binding free energies between different ligands are the typical process in the lead optimization of computer-aided drug discovery. In our study using GPU-TI from AMBER 18 with the AMBER14SB/GAFF1.8 force field, we obtained an overall MUE of 1.17 kcal/mol and an overall RMSE of 1.50 kcal/mol for the 330 perturbations contained in this data set. They are comparable to the overall MUE of 0.9 kcal/mol and RMSE of 1.14 kcal/mol using their GPU free energy code (FEP+) and the OPLS2.1 force field combined with the REST2 enhanced sampling by Schrodinger, Inc. Notably, after we published our work, several other research groups reported their benchmarking results on the other free energy software using the same dataset.The third study of this thesis focuses on modeling the thermodynamics of transition metal (TM) ions binding to a protein. TM ions are very common in biology and are important in drug discovery as well, because many TM ions are in the active site of the protein where the inhibitors bind, for example, the histone deacetylase. While the structural details of TMs bound to metalloproteins are generally well understood via experimental and computational means; studies accurately describing the thermodynamics of TM ion binding are less common. Herein, we demonstrate that we can obtain accurate structural and absolute binding free energies of Co2+ and Ni2+ to the enzyme glyoxalase I (GlxI) using an optimized 12-6-4 (m12-6-4) potential. Optimizing the 12-6-4 potential to accurately model the interactions between the TMs and the binding site residues, as well as protonation state changes associated with TMs (un)binding, are found to be crucial. Given the success of this study, we are now in a position to explore more complicated processes associated with TM-based drug discovery.
Read
