OpWise: Operons aid the identification of differentially expressed genes in bacterial microarray experiments

BMC Bioinformatics

Table 1 Systematic bias in four biological data sets.

	dvSalt30	ecox	shHeat5	shCold5
Typical bias	0.25	0.12	0.37	0.88
Bias/signal (%)	70.4%	19.6%	49.9%	86.9%
Bias/replication error (%)	72.7%	35.8%	143.1%	199.1%
Bias/total (%)	52.4%	15.8%	47.2%	74.6%
Significance of bias
Likelihood ratio	1.74e+02	9.38e+00	1.48e+03	1.81e+03
p-value	< 10^-77	< 10^-5	< 10^-646	< 10^-786

The typical size of the bias in the apparent log2-ratio is the square root of its variance, or $\sqrt{E (1 / θ_{i} \cdot γ))}$ , where E(1/θi) = α/(ν - 1). The bias over the signal is the square root of the ratio of variances ( $\sqrt{β / γ}$ ). The bias over the replicate error is also the square root of the ratio of variances ( $\sqrt{1 / γ}$ ), and considers a single measurement (is not divided by the number of replicates). We also report the typical bias divided by the standard deviation of the observed log-changes mi. To show that the bias is statistically significant, we compared the likelihood ratio of the best-fitting model given systematic error to that without (with γ = ∞), using Eq. 10. Because we are testing whether γ lies at a boundary, in the absence of bias the distribution of 2·log(ratio) approximates a 50:50 mixture of two chi-squared distributions with 0 and 1 degrees of freedom [26].

ISSN: 1471-2105