| Existing computational approaches for protein structure modeling and prediction typically involve a multi-term potential energy function, more commonly referred to as a 'force field'. These force fields typically model various physical phenomena, such as solvation effects, electrostatics, hydrogen bonding, and Lennard-Jones interactions. Occasionally, they also include statistical terms derived from observed protein structures. The combination of these physical and statistical components obfuscates the process of determining and understanding the mechanisms that led to a particular prediction. The variability of force field weights and inclusion of differing empirical data further complicates this analysis.;To address this problem, my dissertation examines the extent to which a simplified energy function that includes only stereochemical constraints and repulsive hard-sphere interactions can correctly predict side-chain dihedral angles of amino acids in protein cores and other complexes. This investigation also investigates the limitations of the hard-sphere model and how to surmount them. My research builds upon the pioneering work of the G. N. Ramachandran, who, in 1963, utilized purely repulsive forces to calculate the allowed cp-ip backbone dihedral angles of dipeptide mimetics. In my work, I apply a similar hard-sphere model to predict the dihedral angles of amino acid side-chains by taking into account both intra- and inter-residue steric interactions.;This dissertation presents three major computational studies. I first examine the hard-sphere predictions of single and multiple repacking of residues in protein cores, and demonstrate that steric interactions are the dominant force in specifying side-chain conformations for several hydrophobic residues. However, these interactions alone are not sufficient to recapitulate the side-chain orientations of methionine residues. I address these limitations in the second study by coupling an attractive force with the hard-sphere model. This addition biases side-chains with more than two dihedral angles towards a more compact conformation, and enables us to recapitulate the observed ensemble side-chain dihedral angle distribution of methionine without affecting the results for other residues. Lastly, I analyze the prediction accuracy of the hard-sphere model as a function of relative solvent accessible surface area, a common metric in the field of structural biology. I show that the hard-sphere model generates correct predictions for monomeric, multimeric, and transmembrane residues with relative surface areas of up to 40%. This emphasizes both the generality of my physics-based model and the behavioral similarity of amino acids from different protein regions and protein types.;This work contributes to the forefront of computational protein design by providing an alternate, simplified approach to protein modeling and an understanding of the limitations of the hard-sphere model. With this information, we can gain fundamental insights into protein structure, specifically on the packing of core residues, which endow proteins and protein complexes with stability. By learning the rules for the successful design of protein cores, we can in turn design new proteins and protein-protein interactions for many potential applications including point of care diagnostics, sensors for proteinaceous biological warfare agents, and more effective vaccines. |
