Using computational alchemy to predict protein-ligand binding free energies
- Author(s): Lawrenz, Morgan
- et al.
Alchemical free energy methods use the computer to make unphysical changes to select atoms of a system during molecular dynamics (MD) simulations, which facilitate computation of the free energy difference between two physically relevant states. For computation of protein- ligand binding free energies, alchemical intermediate simulations interpolate between the protein-bound ligand state and the unbound ligand state. The use of multiple, short, independent MD simulations has been shown to improve conformational space sampling compared to single trajectories of longer length. This dissertation describes the application of this approach to the thermodynamic integration (TI) method for computing binding free energies from alchemical simulations. Particularly in chapters two and four, the independent trajectory TI (IT- TI) approach is demonstrated to give more accurate binding free energy estimates than single TI estimates and allows calculation of a statistically well-defined uncertainty. We use IT-TI to compute accurate absolute. binding free energies for several inhibitors to the avian influenza viral protein neuraminidase and, in chapter four, also to the Mycobacterium tuberculosis enzyme RmlC. The N1 neuraminidase protein has a solvent exposed active site, with many charged residues residing on flexible loops that interact with complementary ligand charge groups. In contrast, RmlC has a smaller, less solvated, and more hydrophobic binding pocket, adding a new test case for the performance of IT-TI. In chapter two, we also use non- alchemical MD simulations to analyze important binding interactions and hydration changes within the N1 active site. We quantify the thermodynamic contributions of these interactions with relative binding free energies computed with IT-TI for alchemical modifications of select ligand moieties. In chapter three, we take further advantage of the dynamic, atomistic description of biomolecular systems afforded by MD simulations to characterize the effect of a calcium ion that binds N1 near the active site. MD simulations, along with IT-TI alchemical free energy calculations, indicate that this bound ion is key for an effective N1 binding pose and is important for accurate models of the N1-drug interface. Current implementations of alchemical methods require many user-defined and non- standardized inputs for performance of these calculations. In chapter four, we use IT-TI on both the N1 and RmlC systems to test varied TI protocols, which we find to have a significant impact on the reliability of free energy estimates. We propose a protocol that maximizes predictive power for the two investigated systems, while allowing speed up of the calculations with distributed computing. Altogether, we show the utility of IT-TI for thermodynamic analysis of biochemical events, and anticipate the application of IT-TI to develop improved and potent drugs