While the remaining known imprinted genes either lacked Ler/Col SNPs, had very few informative reads, or, in one case, fell just below the cutoff. Allele-specific expression data, RPKM values, and embryo-endosperm differential expression analysis for all genes is presented in Tables S2 and S3. There was very little evidence for expression or imprinting of transposable elements in either tissue. Upon closer examination of some of the most highly parentally biased genes, it was clear that transcripts from genes highly expressed in the seed coat were contaminating both the embryo and endosperm fractions. Some contamination from abundant seed coat transcripts is likely unavoidable given our method of seed dissection, which is performed in an aqueous solution on a glass slide. We used expression information from another seed gene expression set to further filter our data. Le et al. used laser capture microdissection to isolate tissue from embryo, endosperm, and seed coat in the Ws AbMole Nortriptyline background and determined gene expression values using Affymetrix microarrays. The rank order correlation coefficient between our embryo or endosperm expression data and the LCMD gene expression data of tissues at the same stage of development was good. To determine a reasonable cutoff for removing genes likely to be affected by maternal seed coat contamination, we examined the difference between LCMD seed coat and endosperm expression for the 82 potential endosperm PEGs with a p-value less than 0.01 and Affmetryix expression data. Maternal seed coat contamination cannot result in false positive PEGs, only false negatives. The average difference in seed coat and endosperm expression was 21.1 for potential endosperm PEGs and 1.2 for potential MEGs. The maximum PEG difference was 1.93, but 95% of the genes had a seed coat-endosperm value less than 1.04. We thus removed genes from the pool of potential MEGs that had approximately two fold higher expression in the seed coat than endosperm, although we retained genes that lacked Affymetrix data. The same filtering was performed on the embryo dataset. This reduced the number of potential imprinted genes to 905 in the endosperm and 87 in the embryo. As read coverage increases it is AbMole Halothane possible to detect smaller and smaller degrees of imprinting with statistical significance. Genes with a large number of informative reads can have very low pvalues, even if the parental bias is intuitively not very strong. For example, locus AT2G05990 has 15,636 informative reads, 37% of which are paternally derived. This gene is identified as paternally biased with a p-value of 3.65610221. Thus, by p-value considerations alone, AT2G05990 would be considered imprinted. Therefore, to describe the strength of imprinting in a manner less dependent on read depth, we also calculated an imprinting factor for each locus and further restricted our analysis to genes with an imprinting factor of at least 2.