Improving statistical models for discovering cell type specific genes

<p>To reveal genetic control of various subsets of CD4 T cells we compared gene expression profiles<br>of resting and activated conventional CD4 T cells, resting and activated natural Treg cells and<br>adaptive Treg cells. RNA was isolated from the respective T cell populations and hybridized to<br>Affymetrix GeneChip M430 2.0 Plus microarrays. Three individual samples of each kind were<br>processed.</p> <p><br>In order to make our data set more representative we included microarrays from the respective<br>CD4 T cell subsets from other laboratories. These data were obtained from the GEO database:<br>www.ncbi.nlm.nih.gov/geo.</p> <p><br>Using the LiMMa framework improved by NMF gene selection we presented ≈ 100 genes divided into 5 clusters<br>specific for corresponding investigated 5 cell types. We pointed out that applying same analysis to data from<br>multiple sources requires extreme care, as differences between sources might have strong influence on results.We<br>proposed a framework (Iterative LiMMa procedure) which allows to reduce, to some extend, source effect in<br>each sample. Nevertheless, the complete solution for removing the effect of the origin remains unknown.