Predicting protein secondary structure with a nearest-neighbor algorithm