Gravar-mail: How Good Are Simplified Models for Protein Structure Prediction?