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

Morgan N Price,Adam P Arkin,Eric J Alm

doi:10.1186/1471-2105-7-19

Abstract

BackgroundDifferentially expressed genes are typically identified by analyzing the variation between replicate measurements. These procedures implicitly assume that there are no systematic errors in the data even though several sources of systematic error are known.ResultsOpWise estimates the amount of systematic error in bacterial microarray data by assuming that genes in the same operon have matching expression patterns. OpWise then performs a Bayesian analysis of a linear model to estimate significance. In simulations, OpWise corrects for systematic error and is robust to deviations from its assumptions. In several bacterial data sets, significant amounts of systematic error are present, and replicate-based approaches overstate the confidence of the changers dramatically, while OpWise does not. Finally, OpWise can identify additional changers by assigning genes higher confidence if they are consistent with other genes in the same operon.ConclusionAlthough microarray data can contain large amounts of systematic error, operons provide an external standard and allow for reasonable estimates of significance. OpWise is available at .

Highlights

Expressed genes are typically identified by analyzing the variation between replicate measurements
We tested the agreement with operons of genes having varying levels of significance. For both two-color cDNA data and Affymetrix oligonucleotide data, our method finds significant amounts of systematic error and reports plausible p-values that show a gradual reduction in agreement with operons as significance decreases
We tested OpWise on four data sets collected with a variety of measurement platforms that used different methods of controlling systematic bias and from several different bacteria: With these data sets, we first used simulations to test whether OpWise fit the data and whether OpWise was robust to deviations

Summary

Results

We tested OpWise on four data sets collected with a variety of measurement platforms (both glass sides and Affymetrix chips) that used different methods of controlling systematic bias (multiple probes per gene or dye swap) and from several different bacteria: With these data sets, we first used simulations to test whether OpWise fit the data and whether OpWise was robust to deviations http://www.biomedcentral.com/1471-2105/7/19 from its assumptions. We compared the operon-wise pvalues to either the ideal or true significance These gave similar slopes as the single-gene p-values, but with consistently smaller deviations from 1.0 (data not shown). For each data set and for three methods of assessing significance (OpWise, OpWise without bias, and significance analysis of microarrays), we divided the changers into eight groups of genes with different levels of confidence. To summarize the performance of the various methods considered here – SAM, single-gene OpWise p-values, operon-wise p-values, and single-gene OpWise with bias ignored – we report the number of putative changers identified at a confidence threshold of 0.05 and the agreement with operons of those changers (Table 2). Genes that are not in operons are included in the operon-wise results

Background

Method

Conclusion

Smyth GK

14. Adhya S