RoDEO: Non-Parametric Robust Differential Expression Operator

RoDEO* is a framework for robust characterization of expression magnitude and detecting differentially expressed genes as well as stable genes between RNA-seq experiments. ( *Robust DE Operator: http://researcher.watson.ibm.com/project/5513 )
For detecting differentially expressed genes, we use a normalization that is not based on the relative values of the gene expression but on the relative order of the expression within a sample. Indeed, the expression values of all genes in an experiment are utilized in a re-sampling approach, to tease out robust relative ranks of the genes in several re-generated instances of the RNA-seq experiments.

Instead of directly working with the expression value of g (such as read count or RPKM), we define a character function φ for each gene g. The two most desirable properties of this function are (i) it depends on the expression values of all the other genes in the assay and and (ii) it is scale invariant. The character function values are estimated by the re-sampling and characterizing the genes into a robust sets of P ranks per sample (e.g. P=20). The character function values of a gene g in case and control experiments are compared to determine whether the gene is differentially expressed.

RoDEO has been applied on reed canary grass (Phalaris arundinacea) cultivars' RNA-seq data to detect differentially expressed and stable genes during salt stress [Comparative Exomics of Phalaris cultivars under salt stress. BMC Genomics 15(Suppl 6):S18, 2014]. Application on water stress is presented in a talk at the PAG XXIII Grasslands workshop.

In this talk, the RoDEO method is explained and current work on additional applications of RoDEO on transcriptome analysis is discussed.