Advances in Force Field Development and Sequence Optimization Methods for Computational Protein Design
<p>The overall goals of computational protein design range from designing new protein folds and protein-protein interfaces to the de novo design of enzymes. All goals require that two equally challenging components of computational protein design be addressed. First, the physical model that de...
| Summary: | <p>The overall goals of computational protein design range from designing new protein folds and protein-protein interfaces to the de novo design of enzymes. All goals require that two equally challenging components of computational protein design be addressed. First, the physical model that describes a protein’s intermolecular and intramolecular interactions must be accurate. Second, energetically optimal amino acid sequences must be identified from an enormous number of possibilities. This thesis describes work that makes progress in both these arenas. In addition, the effectiveness and applicability of computational protein design is demonstrated by tackling challenging design problems.</p>
<p>Improvements to the physical model have been made by developing a more accurate method for calculating rotamer (amino acid side-chain conformation) surface areas for use in our surface area-based hydrophobic solvation term. With this method, surface area errors were decreased dramatically and the experimental stabilities of proteins generated from computationally predicted sequences were improved. Also, our direct surface area calculation approach significantly reduced the compute time required for sequence optimization using dead-end elimination (DEE)-based algorithms.</p>
<p>Although DEE-based algorithms have been effectively used for many challenging design problems, the daunting task of sequence optimization can cause even the most efficient DEE-based methods to fail. We developed a sequence optimization technique called Vegas that combines elements of non-DEE-based as well as DEE-based algorithms. For design problems that were already tractable using DEE-based methods, Vegas delivered the GMEC in significantly less time. In cases where DEE-based algorithms stalled and failed to deliver the GMEC, Vegas produced an answer that, at the time, was better than any other algorithm. This is illustrated by Vegas’ solution to a challenging problem: the full sequence design of a 51-residue fragment of the Drosophila engrailed homeodomain (ENH). We generated a variant of ENH predicted by Vegas and compared its thermodynamic properties with a protein obtained using a Monte Carlo search. We found that the thermodynamic properties of the two molecules were identical. We also solved the solution structure of the Vegas-based molecule using nuclear magnetic resonance (NMR) spectroscopy and found that it folded accurately into the target fold.</p>
<p>Obtaining water soluble variants of membrane proteins might alleviate some of the problems encountered when working with them and facilitate our understanding of the different forces contributing to protein stabilities in membranes. We made progress in developing an automated design scheme that can generate water soluble variants of membrane proteins. We analyzed and compared the surfaces of membrane proteins and water soluble proteins, and developed a metric for altering membrane protein surfaces. Using this metric, we can design membrane protein surfaces using the ORBIT suite of protein design algorithms and convert them to those resembling water soluble protein surfaces. We tested this strategy on two proteins and although we have not been completely successful, we have established rules and guidelines that will aid future efforts towards achieving this goal.</p> |
|---|