Abstract (eng)
The determination of free energy differences is fundamental to the study of several processes such as the binding of drugs to proteins, the paths of enzymatic reactions or the solubility of chemical compounds. By employing molecular dynamics simulations, free energy calculations are capable to compute such free energy differences with high accuracy. However, this accuracy comes at excessive computational costs, often requiring days or weeks to obtain exact results. Thus, considerable effort still has to be invested in the optimization of such techniques.
The first half of this dissertation focuses on the application of Bennett's Acceptance Ratio method (BAR) to problems where standard methods to compute free energy differences are not feasible. This highlights the unique versatility of BAR. Furthermore, we demonstrate how to extend BAR in order to make use of non-Boltzmann probability distributions in biased simulations. We refer to this method as Non-Boltzmann Bennett (NBB). The NBB method is illustrated by several examples that demonstrate how a creative choice of the biased state can also improve the efficiency of free energy simulations.
The second half is concerned with the application of BAR and NBB to the study of hydration free energies. Especially in protein folding or ligand binding (de)-solvation penalties can contribute considerably to the free energy difference. Unfortunately, hydration free energies of amino acids cannot be measured experimentally. Thus, approximations based on side chain analog data are used instead. However, the assumption that side chain analogs are representative for full amino acids has never been thoroughly tested. We, therefore, computed both relative and absolute solvation free energies of amino acids and side chain analogs, showing that the results can deviate considerably due to two effects: Solvent exclusion and self-solvation. While the former accounts for the reduction of solute--solvent interactions due to sterical occlusions, the latter arises from interactions between the backbone and the polar functional groups of the side chains. Since several techniques in computational chemistry do not account for self-solvation, this finding has severe consequences. We illustrate this for several implicit solvent models and briefly discuss the implication of our results for the field of protein science.