The algorithm works by sliding a window along the sequence. Within the window, the minimum free energy of the subsequence is calculated. To evaluate the significance of the local structure signal its minimum free energy is compared to a background distribution of minimum free energies obtained from shuffled sequences, using -scores [Rivas and Eddy, 2000]. The -score statistics corresponds to the number of standard deviations by which the minimum free energy of the original sequence deviates from the average energy of the shuffled sequences. For a given -score, the statistical significance is evaluated as the probability of observing a more extreme -score under the assumption that -scores are normally distributed [Rivas and Eddy, 2000].