Computational simulations of protein-ligand molecular recognition via enhanced samplings, free energy calculations and applications to structure-based drug design

The objective of this dissertation is to understand the underlying intermolecular interactions between disease-causing proteins and anti-disease small-molecule drugs (also known as ligands) through computational methodologies. The insight gained into protein-ligand molecular recognition in terms of structural and energetic complementarities can be rationally utilized for efficient drug discovery, i.e., structure-based drug design (SBDD), to treat diseases caused by specific proteins, including cancers caused by dimerization of survivin (Chapter 2) and of STAT3 (Chapters 4 and 5), and a degenerative, cataract-inducing eye disease caused by misfolding of gammaS-crystallin (Chapter 3).;Protein structures are not static in the body; thus, a single structure cannot fully represent a protein's characteristics. Proteins in aqueous environments have ensemble structures that range from small-scale variations, such as ligand-induced conformational changes (Chapter 2), to large-scale variations, such as protein unfolding and denaturation (Chapter 3). A non-Boltzmann ensemble distribution ('Generalized Boltzmann distribution') can be realized through enhanced sampling methods which help to overcome the pre-existing free energy barriers along the potential energy surface, allowing a wide range of protein structure conformations to be sampled. The extraction of an essential reaction coordinate that constitutes a protein reaction pathway, followed by the construction of a Potential of Mean Force (PMF: the free energy landscape as a function of a selected reaction coordinate), can be utilized for conformational mapping to elucidate the protein's transition structure and thus understand its mechanism.;For candidate ligands to become effective drugs, they should bind to specific protein binding sites with sufficient affinity while controlling the biological reactions of interest. Therefore, estimating the binding energy of a ligand is a crucial metric in assessing a drug's potency. Furthermore, the predicted binding energy should be sufficiently accurate to discriminate the subtle differences associated with its characteristic specificity among similar candidate ligands, as medicinal chemistry usually deals with libraries that contain a large number of chemical compounds. The characterization of the intermolecular forces required for protein-ligand recognition via classical force fields (Chapter 4) and quantum mechanical force fields (Chapter 5) by extracting amino acid residue-by-residue contributions to the ligand binding energy has enriched the quantitative information that can be used for SBDD.