Model development and curve-fittinga.

aConcluding (optimal) models are indicated in bold and their details are presented in other tables and figures.bSM, single (non-bulge) mismatch; BM, bulge mismatch; TM, tandem mismatch; sln, for in solution hybridization; ma, for microarray hybridization.cParameters in italics are used in best-fitting. The use of additional parameters α and β, when applicable, is shown in the free energy column under model description (see text for best-fitting values). Other parameters derived using M3, M5, and M9 are the free energy rules in Tables S1A, S1B, and S1D, with 10, 104, and 8 additions, respectively.dNF, number of probes used in fitting; NT, total number of probes; ε2val and ε2ov, average squares of prediction errors in validation set (i.e., not used in fitting) and overall set, respectively. See Materials and Methods for other parameters.eAbsolute value of the deviation of predicted half-denaturation point (% formamide) from the apparent experimental value, as described by its average/standard deviation (µ/σ) and percentage of values below 5% formamide.fΔG° for perfect matches and ΔΔG° for mismatches (see Materials and Methods).