Computational analysis of protein-ligand binding : from single continuous trajectories to multiple parallel simulations

The interaction of proteins with other proteins or small molecules is essential for biological
functions. Understanding the molecular basis of protein-ligand binding is of
a vast interest for drug discovery, and computational methods to estimate proteinligand
binding are starting to play an increasingly important role. In order to apply
atomistic computational methods to the drug discovery process it is necessary
to have accurate three-dimensional structures of the target protein and a fast and
reliable method to estimate the binding affinity between the target protein and potential
inhibitors. Unfortunately, three-dimensional structures are not available for
all proteins of interest, but often their coordinates can be predicted by computational
methods such as homology modeling.
In this thesis we study the effect of inaccuracies of homology models to ligand binding
using HIV-1 protease as a model system. Homology models of decreasing accuracy
are built and additional errors are introduced by misplacing side chains during
rotamer modeling. We establish a MM-GBSA approach to estimate protein-ligand
binding free energies, and apply this method to the different homology models.
Although MM-GBSA methods are significantly faster than traditional MM-PBSA
methods, still the required computational effort is significant as it is based on the
calculation of a continuous molecular dynamics trajectory. In this study, we establish
a novel approach based on multiple independent short simulations, which is suitable
for execution of a distributed grid of computers and thereby dramatically reduces
the computation time needed. This workflow is validated using the HIV-1 protease
model system, and then applied to the estrogen receptor. Novel methods to assess the
sampling of the different trajectory approaches and potential application to docking
problems are presented and discussed.