Proteins accomplish their functions through conformational changes, often brought about by changes in environmental conditions or ligand binding. Predicting the functional mechanisms of proteins is impossible without a deeper understanding of conformational transitions. Dynamics is the key link between the structure and function of proteins. The protein data bank (PDB) contains multiple structures of the same protein, which have been solved under different conditions, using different experimental methods or in complexes with different ligands. These alternate conformations of the same protein (or similar proteins) can provide important information about what conformational changes take place and how they are brought about. Though there have been multiple computational approaches developed to predict dynamics from structure information, little work has been done to exploit this apparent, but potentially informative, redundancy in the PDB. In this work I bridge this gap by exploring various knowledge-based approaches to understand the structure-dynamics relationship and how it translates into protein function.

First, a novel method for constructing free energy landscapes for conformational changes in proteins is proposed by combining principal motions with knowledge-based potential energies and entropies from coarse-grained models of protein dynamics. Second, an innovative method for computing knowledge-based entropies for proteins using an inverse Boltzmann approach is introduced, similar to the manner in which statistical potentials were previously extracted. We hypothesize that amino acid contact changes observed in the course of conformational changes within a large set of proteins can provide information about local pairwise flexibilities or entropies. By combining this new entropy measure with knowledge-based potential functions, we formulate a knowledge-based free energy (KBF) function that we demonstrate outperforms other statistical potentials in its ability to identify native protein structures embedded with sets of decoys. Third, I apply the methods developed above in collaboration with experimentalists to understand the molecular mechanisms of conformational changes in several protein systems including cadherins and membrane transporters.

This work introduces several ways that the huge data in the PDB can be utilized to understand the underlying principles behind the structure-dynamics-function relationships of proteins. Results from this work have several important applications in structural bioinformatics such as structure prediction, molecular docking, protein engineering and design. In particular, the new KBFs developed in this dissertation have immediate applications in emerging topics such as prediction of 3D structure from coevolving residues in sequence alignments as well as in identifying the phenotypic effects of mutants.