COMPUTATIONAL METHODS & FORCEFIELDS FOR PROTEIN DESIGN, STRUCTURE PREDICTION, & REFINEMENT WITH NATURAL & MODIFIED AMINO ACIDS

Abstract

Most methods for modeling, simulating, and designing proteins focus on accurately modeling the 20 natural amino acids. Post-translational modifications (PTMs) and non-canonical amino acids (NCAAs) are chemical modifications of the 20 amino acids and offer an expanded chemical and property space. Accurate models of these molecules in the condensed phase have not been developed. To address this challenge, we have developed two new atomistic forcefields to enable accurate calculations in the condensed phase of over 180 modified amino acids. The statistics of PTMs contained in the Swiss-Prot database were curated to identify those most frequently occurring. We developed Forcefield PTM, a set of AMBER charge and torsion forcefield parameters consistent with ff03 for 32 common PTMs. The parameterization methodology was extended to create charge parameters for 147 NCAAs. The forcefields were validated against experimental hydration free energies through thermodynamic integration calculations in both the TIP3P and TIP4P-Ew water models finding comparable correlations and errors to ff03 for natural amino acids. The forcefields were integrated into a new protein design method that combines integer linear optimization and molecular simulations, and they were used to help design several new analogs of Compstatin. Several analogs exhibited enhanced activity and/or solubility compared to the natural amino acid-containing starting compound. Due to their potency and solubility, these peptides are promising candidates for therapeutic development in complement-mediated diseases.

In the second theme of this dissertation, we developed methods for protein structure prediction and refinement. We describe WeFold, a “coopetition” where 13 labs worldwide strategically combined methods and resources into new hybrid pipelines to determine whether they can outperform their base components. Princeton TIGRESS was created to address the structure refinement problem by separately performing sampling and selection stages. Among refinement methods implemented as web-servers, Princeton TIGRESS was found to demonstrate the most consistent and most substantial net refinement in CASP11. Four webtools have been created to disseminate this research to a broader audience. The methods developed are actively being used to discover new inhibitors and refine protein structures.