Additional statistical testsa,b.

aSee Table 2 for the definition of models and parameters and the reference values. See Text S1 for the details of the statistical tests.bRandomization and permutation tests are based on at least 100 runs until convergence. Significant figures in these results reflect the uncertainty in the converged values.cProbes were permuted while maintaining the original sequence of each probe. This test corresponds to permuting the probe length in addition to the nearest neighbor free energies.dShow improvement over original models although statistical parameters indicate otherwise. The discrepancy reflects the fact that half-denaturation point is not a perfect representation of the melting point for experimental profiles without a plateau (e.g., see Figure S3C). This adversely affects the results with γ-factors more than without, as the models without γ-factors tend to compensate for the lack of good fitting in the vertical by moving closer to experimental values in the horizontal, although this movement does not mean a better match. Since the original models in the main text always use γ-factors, the evaluation of model predictions are conservative and more accurate with respect to half-denaturation points. This analysis provides just another way of seeing how γ-factors buffer experimental artifacts as discussed in Text S1.eBest-fitted γ replaced by a model-derived factor (see Text S1).