|
Abstract:
|
Shrinkage based methods have been promising for inference on differential expression in small microarray experiments. Some experiments are characterized by both biological and technical replicates, the latter of which may include several spots per probe on an array. A popular R package, LIMMA, accommodates this issue by assuming a constant correlation for the within-array replicates on each probe, based on the duplicate correlation (dupcor) option. We assert with an example that this assumption may not be true for many experiments and propose a technique that models heterogeneity on within-array correlations between technical replicates across genes. Based on simulation studies, our technique outperformed the dupcor option of LIMMA for ROC and FDR characteristics, particularly when each gene was spotted more than twice on an array and within-array correlations were high.
|
