Structural characterization of the protein native state is an important problem in computational biology. Thermodynamically, the native state is that of lowest free energy in the protein conformational space . Predicting it ab initio from the amino-acid sequence can be posed as an optimization problem that has proven to be NP-hard . Due to imperfect modeling of interatomic interactions, the native state often does not correspond to the global minimum. As a result, the goal in ab-initio protein structure prediction is to first arrive at a diverse ensemble of low-energy (decoy) conformations potentially relevant for the native state. Decoys are often computed using a coarse-grained energy function that expedites sampling of low-energy conformations. Select decoys are then refined with heavy-duty protocols using fine-grained energy functions to allow prediction of the native state.