Biology Reference
In-Depth Information
replicate pointwise uncertainties were explicitly considered. This
variable allows for employing either (1) pointwise SEMs in variable
weighting for statistical considerations, or (2) no weighting, assuming, in
effect, each data point to be known with infinite precision (i.e., with
associated pointwise SEMs of zero).
Tables 12-4 through 12-8 summarize the results obtained from
simulations with each of the aforementioned analytical strategies as
a function of simulation condition. As expected, higher signal-to-noise
ratios and/or higher numbers of replicates yield more accurate and
more statistically significant results. In the tables, values reported are
means from analyses of the 1000 surrogate time series for each condition
tested. Associated standard deviations appear in parentheses
immediately below the mean values. The column headings have the
following meanings:
SEM refers to whether (Yes) or not (No) individual pointwise
replicate uncertainties were considered during statistical analysis;
S/N denotes signal-to-noise ratio;
% refers to percentage of files identified as circadianly rhythmic (i.e.,
with periods such that 20 hrs
< t <
28 hrs);
MeanExpLev denotes mean expression level of the identified time
series (theoretically zero); values in parentheses are SDs;
Period denotes the period of the identified time series (theoretically
24);
Phase denotes time of acrophase of the identified time series (theo-
retically zero);
Ampl denotes oscillatory amplitude of the identified time series
(theoretically either 2, 1, or 0);
Prob denotes uncorrected significance probability of the identified
time series;
Prob(MMC) denotes multiple measures corrected significance prob-
ability of the identified time series. It is considered circadianly
rhythmic if Prob(MMC)
<
0.05; and
N/D means the value was not determined.
In conclusion, despite the considerable challenges for circadian analysis
of gene chip-derived time series, the COSOPT algorithm appears to
perform admirably. It extracts reliable estimates of period, phase, and
oscillatory amplitude. It requires no user initialization, is stable, and
appears to produce unbiased parameter value estimates. It is readily
