Cryoem-Guided Iterative Molecular Dynamics - Rosetta Protein Structure Refinement Protocol Improves Protein Model Quality

Abstract

There's a huge gap between the sequence and structural knowledge of proteins. Determining proteins structures experimentally is expensive and not always amenable. Computational tools such as homology modeling have been successful at bridging this gap. However, they are still subject to database information and they don't have a principled way of selecting structures. This and other issues are treated in the blind CASP competition (Critical Assessment of Structure Prediction). We took part in the recent CASP11 competition by using a new physics based approach called MELD[1,2] and combining with either proposed data during CASP or with general insights coming from our knowledge of globular protein structures (e.g. they have hydrophobic cores). Physics based methods are usually too computationally demanding for the tight deadlines in CASP, which has mostly excluded them from participation. In this talk I will show how MELD was useful during CASP. In some cases MELD provided value beyond what database methods could do-including the best blind prediction for one of the proteins in CASP. As such, physics based methods are starting to be useful for structure prediction to complement existing methods. 1.MacCallum, J. L., Perez, A. & Dill, K. Determining protein structures by combining semireliable data with atomistic physical models by Bayesian inference.