Clustering of 86 RNA-seq experiments from 5 different studies with 11 different sample types. Heatmaps show hierarchical clustering of the samples based on Pearson correlation of the genes with sliding DFI ranges. When top gene features (with DFI value larger than 3.5) are used, sample type drives the clustering (top-left). As the DFI values get smaller, correlation of the samples get weaker. When genes with DFI values less than 1.5 are considered all of the samples correlate to each other. Series of these heatmaps confirms the differentiation power of the DFI ranking in a single step.
Ozer et al. BMC Genomics 2012 13(Suppl 8):S11 doi:10.1186/1471-2164-13-S8-S11