I have some reasonable idea of what directional means are in the context of spatial statistics, but I am stumped by this use of the term in the methods section of this computational biology paper (emphasis mine):

The raw ES values were normalized to account for variable numbers of shRNAs across different genes by dividing the raw ES by the directional mean of a size-matched null distribution generated by 100,000 random permutations of a hairpin set of the same size.

To sum up the relevant part: one is supposed to compute the directional mean of a distribution obtained from a set of scores (no vectors involved).

How would one go about getting a directional mean from a set of values (or associated density function)?

Edit: one piece of information that could be relevant: the score values whose distribution is being plotted and whose "directional mean" is supposed to be extracted, are obtained through a goodness-of-fit test applied to a subset of values ("the hairpin set") out of the whole ranked list (the microarray data, converted to a list of differential expression values).

I also doubt the spatial aspect of the microarray data could play any role here (it's long been eliminated through pre-treatment).

  • $\begingroup$ Just a guess, but "hairpin set" sounds like it could be directional. $\endgroup$ – Peter Flom Jul 20 '12 at 10:13
  • $\begingroup$ I looked at the article. Although I didn't catch the specific sentence that you mentioned the data was collected on a microarray which does have a spatial aspect to it. I think Peter's guess is correct. $\endgroup$ – Michael Chernick Jul 20 '12 at 10:38
  • $\begingroup$ Thanks for your comment @Peter, but "hairpin sets" here only represent a subset of probes (i.e. each key to a value in the microarray). Although you could say there's a spatial aspect to microarray, it is dealt with (and eliminated) long before, during pre-treatment of the data. I really don't think this spatiality would pop up at this stage, especially when considering a distribution of scores obtained from these (the scores are essentially a goodness-of-fit test on the subset compared to the whole array, in a ranked list of values). $\endgroup$ – Dave Jul 21 '12 at 1:40
  • 2
    $\begingroup$ From my experience the best way to be certain is to email the authors $\endgroup$ – Bitwise Sep 28 '12 at 18:08
  • $\begingroup$ 'directional mean' appears to be a term from spatial statistics; it's unclear if that's what's intended here. $\endgroup$ – Glen_b Jun 26 '13 at 8:41

I think it is not exactly standard terminology, hence it can mean pretty much anything. However, I have found this article which seems to be in the same genre: http://www.plosgenetics.org/article/info%3Adoi%2F10.1371%2Fjournal.pgen.1002621

This makes me suspect that directional may have been used instead of conditional.

But of course directional mean could just be mean, or absolute mean.

  • $\begingroup$ You could perhaps try backtracking the use of the word by checking the sources if there are any known ones? $\endgroup$ – Dennis Jaheruddin Jul 30 '12 at 15:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.