From: Robin Colgrove <robin_at_hms.harvard.edu>

Date: Tue 07 Jun 2005 - 01:13:53 EST

R-help@stat.math.ethz.ch mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Tue Jun 07 01:36:12 2005

Date: Tue 07 Jun 2005 - 01:13:53 EST

Hello all,

Sorry if this is an FAQ. I have been trying to search the archives
without success.

I have a dataset (ChiPs microarray) where the experiment to experiment
variability is very high

but where within an experiment, the data nearly always goes in the
"right" (hypothesis confirming) direction.
I am trying to figure out the right way to use R to do the statistical
analysis and generate an appropriate figure.

To be specific, we have a virus and a mutant derivative, and the hypothesis is that the wild type virus is specifically suppressing transcription activity in a manner that is abrogated in the mutant.

The experiment is to measure the amount of viral chromatin associated with overall histone (should be the same between wild type and mutant), vs. transcriptionally active chromatin (mutant should be greater than wild type) vs. inactive chromatin (wild type should be greater than mutant).

For each datapoint there four variables: a histone type (general, active, inactive), a specific gene assayed (four different genes), a virus used for infection (wild type or mutant), and an experiment number (each combination repeated 3-5 times) These are hard experiments to do (involving dissecting out small numbers of cells from a mouse) so the numbers are small, but in each case, there are 3-5 pairs of wild type vs mutant virus for each condition.

If I look at simply whether the hypothesis is confirmed for each condition (whether the wild type/mutant difference goes the way you would expect), then the sign is right 34/35 times, which is way beyond reasonable significance. However, since the inter-experiment variance is so high, if I try to do a simple rank-sum test for a particular chromatin-gene-virus combination (3-5 pairs), the result is usually non-significant (never significant if Bonferroni corrections for multiple tests are applied).

My questions are:

- Is there a good way within R to do and report a simple sign test on this sort of data (paired samples, non-normal, high-inter experiment variability)?
- What would be a good way to plot this sort of noisy data (and how to do it in R)? I was thinking of having each (wildtype-mutant) experiment pair as the ends of line segments with different colors or line-types for each gene-chromatin type combination. I know this is a standard kind of plot but I can't figure out how to do it in R.
- What is the best way to input this data? A 4d array with virus type (wild type or mutant) on one axis, chromatin type (non-specific, active, inactive) on the second, gene (one of four different genes) on the third, and experiment number (1-5) on the last? Is there a good way to do this with data frames?

Thanks for any help or pointers to appropriate how-to's. I am really trying to figure this out myself, but as a virologist/bioinformaticist new to R, I still have a lot to learn statistics-wise.

robin colgrove

dept. of microbiology

harvard medical school

R-help@stat.math.ethz.ch mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Tue Jun 07 01:36:12 2005

*
This archive was generated by hypermail 2.1.8
: Fri 03 Mar 2006 - 03:32:23 EST
*