Method development and application in protein structure prediction, computational molecular docking and network biology

The sequence of the human and other genomes has provided us a list of parts (e.g., proteins) in the molecular machine of life. However, the biological functions for most of these "parts" remain unknown. Moreover, our knowledge of how these individual parts interact with each other and how these interactions work together as a molecular network in living cells is very limited. In this dissertation, I develop and apply novel bioinformatics methods to address these questions.; In part I, I report new methods for predicting the three-dimensional (3D) structures of proteins. Protein structures are intrinsically related to their respective functions. Chapter 2 describes an algorithm that recognizes the structural fold of a query sequence from the known structures of other proteins. In this new algorithm, the predicted solvent-accessibility profile is employed to capture the local pattern of 3D packing. Chapter 3 presents a method for modeling the unconserved, structure variable regions ( e.g., loops). The method combines an efficient statistical energy function with a side-chain optimization protocol. In chapter 4, the performances of different simplified energy functions are benchmarked for native structure discrimination.; In part II, I present new docking techniques for predicting the three-dimensional structures of protein complexes. In chapter 5, an atomic potential of mean force derived from high-resolution protein-monomer structures is used to calculate the free energy of protein-protein binding. This energy function gives an accurate prediction of protein-protein binding affinities. In chapter 6, the energy function is further extended to study protein-ligand and protein-DNA interactions. In chapter 7, I establish a local docking pipeline which integrates a scoring function, a sampling program and a clustering algorithm to predict the structures of protein complexes using independently solved monomeric structures. The method is tested in CAPRI (Critical Assessment of PRediction of Interactions) for blind prediction of the structures of protein-protein complexes. In chapter 8, I developed a method for binding site prediction of monomeric proteins. The method linearly combines free energy score, evolutionary conservation score and interface-propensity score to predict a continuous binding region in protein surface.; In part III, I describe various methods to explore the molecular network within living cells. (Abstract shortened by UMI.)...