A combinatorial distance-constraint approach to predicting protein tertiary models from known secondary structure