Table 1. Linear regression parameters for the model in equation 3 for the PM data.

PM	Intercept	γA	γC	γG	R2
ln a	6.617 ± 0.167	0.008 ± 0.014	0.219 ± 0.014	0.195 ± 0.013	0.56
ln b	0.768 ± 0.324	0.154 ± 0.022	0.206 ± 0.028	0.377 ± 0.026	0.44
ln d	2.533 ± 0.416	–0.305 ± 0.028	0.354 ± 0.035	0.168 ± 0.033	0.48

Most parameters have small standard errors compared to their values, indicating that the fits truly capture sequence specificity. Probabilities p(γ = 0) < 10^–6 under the hypothesis of no sequence-specificity, except for γ^a_A. Probes were excluded from the fit according to the following criteria: (i) (a, b, d) had to be strictly positive because of the logarithms; (ii) an upper limit on b < 10 000 excluded probes in which no saturation effects were observed and hence a and b could not be determined independently; (iii) d < a / 5 excluded probes that were probably subject to significant cross-hybridization; and (iv) the calibration curves had to follow good Langmuir isotherms: the correlation coefficient ρ(ln I_obs, ln I_fit) between the observed and fitted intensities had to be >0.99. In total, this procedure removed 29.7% of the probes.