J.P.Townsend, D.L.Hartl, Bayesian analysis of gene expression levels: statistical quantification of relative mRNA level across multiple strains or treatments, (Genome Biology 3: research0071.1-0071.16;2002).

To assign P-values to differences in male vs. female gene expression in the Drosophila data sets, we used the Bayesian method of Townsend and Hartl (2002), which is implemented in the software BAGEL. This method is quite flexible and can be applied to data sets that come from different platforms and that use different replication schemes. For details of how BAGEL was applied to a specific data set, please see the individual reference page for that microarray data set. In all cases, we used the default settings for BAGEL (Unix version 3.6), including the assumption of additive variance.

The P-values can generally be interpreted as the probability of a type 1 error. In this case, the probability that there is no difference in expression level between males and females. However, there are two important things to note:
1. These P-values are not corrected for multiple testing.
2. Because the different microarray experiments differ in number of genes, number of replicates, and the variance among replicates, the P-values cannot be compared directly across experiments.

To get around these problems, we randomly re-sampled the data in each data set and repeated the BAGEL analysis. This allowed us to estimate the False Discovery Rate (FDR) corresponding to a particular P-value for each data set.

The following table gives the P-value from each data set that corresponds to an FDR of 10%. For example, if one uses a P-value cutoff of 0.025 to classify genes from the Ranz et al. (2003) D. melanogaster data set as sex-biased, then one would expect that 10% of these genes are false positives (i.e. they really show equal expression in the two sexes).

Data set p-value (FDR 10%)
Ranz(2003) 0.0250
Parisi(2004)[testes/ovaries] 0.0150
Parisi(2004)[male/female] 0.0425
Parisi(2004)[gonadectomized] 0.0500
Gibson(2004)[Ore-R] 0.0350
Gibson(2004)[2b] 0.0250
Gibson(2004)[Ore-R + 2b] 0.0225
McIntyre[Ore-R] 0.0250
McIntyre[2b] 0.0200
McIntyre[Ore-R + 2b] 0.0300
Goldman(2007) 0.0200
Meta-analysis(v2.0; 2009) 0.0265
Innocenti(2010) 0.0090
Wyman(2010) 0.025
Meta-analysis(v3.0; 2011) 0.010
Data set p-value (FDR 10%)
Ranz(2003) 0.0375
Note: In Sebida, a P-value of '0' indicates that P < 0.0001. If the field is blank, that means that not enough data were available for a P-value calculation.