More accurate comparative (homology) models would enable greater biological understanding through structural genomics efforts as well as aid in biological and small-molecule drug development. However, improving the accuracy of comparative models beyond that of the homologous template protein has proven extremely difficult for many years. The primary aim of this work is to develop more accurate, molecular mechanics-based, computational methods for refining loops in comparative models. My approach is two-fold:

A. Create a set of protein "model systems" that exhibit specific types of modeling error as found surrounding loops in comparative models

B. Develop new loop-sampling methods that optimize atoms outside the loop

The types of structural errors found in comparative models can be divided into bins based on sampling degrees-of-freedom: 1. side-chain error, 2. backbone error, and 3. larger-scale structural errors, such as helix or domain orientations. In chapter 1, we perturbed crystal structures to contain side-chain errors exclusively (error type 1.) We then augmented our previous loop prediction method to simultaneously optimize side-chains surrounding the loop. Results show that our new method can recover the native state in most of the cases where our previous method failed. In chapter 2, we chose homology models of antibodies as a test system to investigate loop prediction when the surrounding backbone atoms are incorrect (error type 2.) We predict the antibody H3 hyper-variable loop ab initio while the remaining five, hyper-variable loops are modeled using loop templates whose backbone atoms tend to deviate slightly from native. By increasing H3 loop sampling and performing optimizations on the surrounding loops iteratively, we were able to increase accuracy over previous methods. In chapter 3, through collaboration with Arjun Narayanan, we have taken initial steps in analyzing the determinants of variation in antibody light and heavy domain orientation (error type 3.) In chapter 4, through collaboration with Sergio Wong, I applied these new loop prediction methods to investigate a previous hypothesis that antibody H3 loops rigidify during affinity maturation which occurs within B-cells upon antigen encounter. In summary, this work constitutes a significant step towards a general method of comparative model refinement.