The main goal of this proposal is to produce high-accuracy comparative models for proteins that share less than 30% sequence identity with their homologs of known structures. Model accuracy is influenced by the target-template sequence alignment accuracy and the structural similarity between target and template. We work under the assumption that model inaccuracies that have different origins (misalignment vs. structural divergence) need to be dealt with in different ways, and that effective improvement of model accuracy requires understanding the nature of the interplay between """"""""alignment error"""""""" and """"""""template error."""""""" In this project the alignment error is dealt with through a new alignment optimization procedure guided by structure-based evaluation of the model implied by the alignment. The template error is dealt with a new refinement approach that combines simulation with template-derived restraints. Molecular dynamics and Monte Carlo simulations are used to explore conformational space and to provide an accurate refinement environment that includes explicit solvent. Finally, alignment error and template error are iteratively optimized in an approach that exploits the synergy of alignment and structure refinement. To guarantee that the approach is practical it is also applied to a real modeling project with experimental validation. This multidisciplinary approach is possible because of the combination of researchers in comparative modeling (Dr. Sanchez), computational biophysics (Drs. Osman and Mezei), and experimental structural biology (Dr. Zhou). At the end of the project period it is expected that a new modular approach to comparative modeling will have been developed that is capable of modeling remote homologs of known structures with an accuracy equivalent to that of high-resolution NMR structures.
