Objectives: Single genes are not, in general, the primary focus of gene expression experiments. The researcher might be more interested in relevant pathways, functional sets, or genomic regions consisting of several genes. Efficient statistical tools to handle this task are of interest to research of biology and medicine.
Methods: A simultaneous test on phenotype main effect and gene-phenotype interaction in a two-way layout linear model is introduced as a global test on differential expression for gene groups. Its statistical properties are compared with those of the global test for groups of genes by Goeman et al. in a preliminary simulation study. The procedure presented also allows adjusting for covariates.
Results: The proposed ANCOVA global test is equivalent to Goeman's global test in a setting of independent genes. In our simulation setting for correlated genes, both tests lose power, however with a stronger loss for Goeman's test. Especially in cases where the asymptotic distribution cannot be used, the stratified use of the ANCOVA global test shows a better performance than Goeman's test.
Conclusions: Our ANCOVA-based approach is a competitive alternative to Goeman's global test in assessing differential gene expression between groups. It can be extended and generalized in several ways by a modification of the projection matrix.