**Half-life calculation
by non-linear least squares curvefitting**

In order to
improve the statistical rigor and accuracy of the half-life calculation, the
data for each RNA was fitted to an exponential function (of the form A = A_{0}e^{kt})
using a non-linear least squares algorithm implented in MATLAB (function nlinfit
in the statistics toolbox which uses the Gauss-Newton method). The function nlparci was then used to
estimate 95% confidence intervals for the two parameters (A_{0,low}, k_{low},
A_{0,hi}, and k_{hi}) which were then used to calculate upper
and lower bounds of the half lives.
Half-lives were calculated for 2,679 RNAs for which A_{0,low}
was positive (95% confidence of detection) and k_{hi} was negative (95%
confidence of decrease). Half-lives are
reported in minutes. 907 half-lives
calculated by both methods are plotted in figure 1.

Transcripts which fall along the line fit an exponential degradation pattern, whereas those that fall off the line do not. Those above the line degrade more slowly at the beginning of the timecourse (over their first 2-fold change) and more quickly later on. Those below the line have the opposite pattern.