The design has more than two conditions and samples are independent

If you have more than two conditions, you can compare them pairwise which raises serious statistical difficulties. Instead, use the multi-condition equivalent of t-scores: F-scores. In R, use mt.teststat(. , test="f") in package multtest. The Wilcoxon equivalent are Kruskal-Wallis scores (kruskal. test). Note that multi-condition scores are sensitive to every gene whose expression is different in any one of the conditions. If the average expression in one condition deviates strongly from all others, the score is high.

A special case in multi-condition settings are dependent samples. Consider a set of 10 patients, each measured at five distinct time-points. This is a typical design in time-series or survival analysis. There are scores specialized for both, but this is beyond the scope of this chapter.

0 0

Post a comment