A constraint-based approach to structure prediction for simplified protein models that outperforms other existing methods

Lattice protein models are used for hierarchical approaches to protein structure prediction, as well as for investigating general principles of protein folding. So far, one has the problem that either the lattice does not model real protein conformations with good quality, or there is no efficient method known for finding native conformations. We present a constraint-based method that largely improves this situation. It outperforms all existing approaches in lattice protein folding on the type of model we have chosen (namely the HP-model by Lau and Dill[34], which models the important aspect of hydrophobicity). It is the only exact method that has been applied to two different lattices. Furthermore, it is the only exact method for the face-centered cubic lattice. This lattice is important since it has been shown [38] that the FCC lattice can model real protein conformations with coordinate root mean square deviation below 2 Å. Our method uses a constraint-based approach. It works by first calculating maximally compact sets of points (hydrophobic cores), and then threading the given HP-sequence to the hydrophobic cores such that the core is occupied by H-monomers.