Skip to main content
. 2007 Apr 19;8:105. doi: 10.1186/1471-2164-8-105

Table 1.

Results of logistic regression for intensity dependence.

Median MAD
dataset original re-loess original re-loess

9a 184 (8.23e-39) 4.37 (0.358) 81.5 (8.28e-17) 62.6 (8.26e-13)
9b 246 (5.02e-52) 3.2 (0.525) 48.1 (9.06e-10) 49.9 (3.79e-10)
9c 225 (1.56e-47) 3.41 (0.492) 83.4 (3.26e-17) 62.3 (9.6e-13)
9d 271 (1.85e-57) 4.02 (0.403) 71.7 (9.93e-15) 59 (4.73e-12)
9e 104 (1.28e-21) 7.61 (0.107) 24.2 (7.38e-05) 45.3 (3.37e-09)
10a 151 (1e-31) 6.61 (0.158) 82.3 (5.69e-17) 35.6 (3.47e-07)
10b 190 (4.86e-40) 3.19 (0.527) 102 (4.63e-21) 32.5 (1.54e-06)
10c 214 (4.52e-45) 8.12 (0.0874) 124 (6.76e-26) 47.7 (1.11e-09)
10d 238 (2.1e-50) 4.62 (0.329) 157 (6.06e-33) 39.4 (5.63e-08)
10e 105 (8.49e-22) 12 (0.0171) 21.9 (0.000208) 36.4 (2.43e-07)

The probability that the control samples will have a value greater than the matched spike-in samples was modeled as the logit of a function of a 4th order polynomial for rankit intensity. Values for the median and the MAD (median absolute deviation) were considered. The deviances and p-values (in bold) for the comparison of the polynomial model to a constant null model are provided and are consistent with the results presented in Figure 6. Re-loessing the data using only the fold change 1 all but eliminates the relationships between intensity and relative centering of the two sample populations. However, the relationships between intensity and relative variability of the expression values remain, although they are greatly diminished.