Identifying genes involved in the growth of adrenocortical tumors: An approach based on partial linear models

Many diseases exist for which some numerical severity indicator can be measured. To identify the differentially expressed genes involved in the progression of such a condition, we propose a method that is more informative than traditional analyses because it accounts for severity measurements in addition to comparing diseased vs. control samples. Our approach incorporates the effects of disease severity as a nonlinear term which is added to a linear model to explain the variation in each gene’s expression level between normal and diseased samples. Applying this method to microarray data from human adrenocortical carcinomas, where expression profiles of each tumor are accompanied by mitotic rate measurements, we report interesting findings on the relationship between gene expression and mitotic rates in adrenocortical cancer. These models have been implemented in the R package plmDE, which facilitates their flexible application to differential gene expression analyses containing information on relevant quantitative phenotypic variables.