Computing Rma Measure - Median Polish

Question

[dirigible2012]

I am trying to code my own routines to calculate the RMA measure similar to RMA express or the affy Bioconductor package. I am having trouble getting my numbers to come out the same as the results from RMA Express.

I am currently working on the median polish step, so I am running RMA Express with NO background correction or quantile normalisation. When I run RMA Express with only one .CEL file, I would expect median polish to return the median for each probeset. This is not the case. i.e.

When I run RMA Express on a probeset with mean values (from the CEL file) 228 112 40 53 and standard deviations 31.3 15.1 5.2 8.5, RMA Express returns the result 77.045441 where I would expect RMA to return 82.5 (the median).

Could somebody tell me how RMA Express is getting this number?

Thanks, Stephanie

*Edited to make things clearer.

[Neilfws]

No-one can tell what is wrong without seeing both your code and the data on which it was run.

[dirigible2012]

Sorry, should have been more clear - I don't need my code fixing, I need to know the algorithm RMA Express is using so I know what to program. Currently it's not behaving as I expected from reading the documentation.

[Neilfws]

I'm not suggesting that your code needs to be fixed. My point is that we cannot reproduce what you are seeing without access to the same CEL file. Telling us mean values, SDs and medians does not help.

Answer 1

[dirigible2012]

Answering my own question: :-) The median polish algorithm takes log 2 of the CEL file values before starting the polish process, then (if you choose the 'Natural Scale' output) converts the answer back to natural scale by raising 2 to that power.

e.g. log 2 of 228 112 40 53 is 7.83289 6.807355 5.321928 5.72792

median of this is 6.267638

2 raised to this power is 77.04544