Major Professors: Dan Nettleton and Roger Wise
Home Department: Department of Statistics
Title of Defense: SPADE: shrinkage and parametric bootstrap differential expression analysis for correlated RNA-seq
Differential expression (DE) analysis is a fundamental task in RNA-seq analysis. Several methods exist for identifying DE genes under independent response assumption, but far fewer are available for identifying DE genes when read counts within a gene are correlated. We proposed a method called SPADE, an acronym for "shrinkage and parametric bootstrap differential expression analysis". In SPADE, we take advantage of the voom-limma approach, transferring the read counts to logarithm counts per million and getting observational specific weights from the mean-variance trend. Then, we fit a general linear model to transformed read counts of each gene with unstructured correlation matrix. We shrink the residual maximum likelihood estimators of correlation parameters toward a mean-correlation trend estimated from all genes. Parametric bootstrap is applied to approximate the null distribution of test statistics.