Adaptive Gene Picking with Microarray Data: Detecting Important Low Abundance Signals

DNA microarrays to evaluate gene expression present tremendous opportunities for understanding complex biological processes. However, important genes, such as transcription factors and receptors, are expressed at low levels, potentially leading to negative values after adjusting for background. These low-abundance transcripts have previously been ignored or handled in an ad hoc way. We describe a method that analyzes genes with low expression using normal scores and robustly adapts to changing variability across average expression levels. This approach can be the basis for clustering and other exploratory methods. Our algorithm also assigns p-values that are sensitive to changes in variability with gene expression. Together, these two features expand the repertoire of genes that can be analyzed with DNA arrays.