Predictions of protein segments with the same amino acid sequence and different secondary structure: a benchmark for predictive methods