Currently I am using a proteomics dataset to compare 4 different disorders to healthy controls. Ultimately, I want to find unique and overlapping factors (that are deviant from healthy controls) in these disorders. The resulting data are thus representable as a four way venn-diagram. However, as any set of 4 lists would show overlaps and unique factors, I find it difficult to judge the amount of 'uniqueness' in comparison to other possible classification with these 4 sets of patients. Therefor I wanted to permute my data into 4 random sets and calculate the number of unique and overlapping factors. After several iterations I would then be able to calculate a p-value starting from the histogram. My question here is two-fold: 1) Is this valid as a meta-statistic i.e. to answer the question if the labels are adequate representations of the underlying biological differences. 2) While I know that some packages facilitate permutation, I just can't wrap my head around how to implement it, due to 2 main issues: - My experimental groups are unbalanced (5-4-4-3) - For the distribution to be valid, it should be done without replacement. However, as the individual groupings themselves lose meaning as soon as the permutation procedure starts and thus there should only be half op the combinations as expected by just excluding replacement within groups. (the groups of 4 are equivalent) I would be very grateful if someone could give me some tips or point me towards interesting packages/reading to help me tackle the problem. I only have moderate experience with R and my preference is to implement it in R but I am open to suggestions.