An additional concern is that current gene-set analysis methods generally use a permutation-based approach. These are often very computationally demanding, and since no parametric model is used it is often not made explicit which null hypothesis is being evaluated and what assumptions are made. This makes it more difficult to determine the properties of the analysis such as how the significance of a gene set relates to the significance of its constituent genes or whether the analysis corrects for a polygenic architecture. This complicates the interpretation of results and hampers comparison between results of different gene-set analysis methods.
