Try free FlexArray software

Hi

I think you can use dCHIP software for this purpose. You can analyse and compare different data sets that was obtained from different platforms

If I understood correctly, you would have a problem with probe IDs at the first level because you have a custom chip that you want to compare with Affymetrix data. Try to see in Genevestigator whether both experiments are there and then export them as CSV or Tab delimited file containing common Uniprot ID and re-align them in Excel. Those without commoon ID discard and make stats on leftovers...

Sorry but which set is non-affymetrix?

You want to compare two different array platforms? if you have a common reference you can try to normalize but the probes may not respond in a similar way. Abs present calls are just thresholds based on signal, You can approximate this typically by using 2-3 fold over the global median expression level of the array. So if a probe signal is 2-3 fold over the global median you can call it present if it is under call it absent. Crude but it tends to work. You can empirically test this on a sample of probes that you know are expressed and not expressed. You do not need a package you can do this in R or even excel Mu11 has only 11K probes. Not sure how many arrays you are using though.

I would not recommend comparing intensities between two platforms in a genome-wide analysis. Each reporter on each platform should be thought of as separate assays that attempt to measure concentration of RNA. However the platform does not report concentration, it reports an intensity value that may be proportional to concentration over some undefined concentration range. There is no reason to think that, and it would be dangerous to assume that, the linear ranges are the same between batches from a single platform much less between platforms or reporters. The problem occurs at high or low concentrations that yield very limited information in one or both platforms because the concentration is beyond the upper or lower limits of detection (or quantification). This information cannot be retrieved by normalization, and analysis of such genes can definitely lead to statistical artifacts.

Instead of combining the data, I suggest analyzing each dataset separately and doing a meta-analysis on the results of that (i.e. one becomes a test and the second a confirmation). I would also recommend examining absolute intensities to see if they might be affected by upper and lower limits.

Michael's warnings and approach are reasonable, but only if you have samples A and B and they have sample A and B. What is the data and what are you trying to find?

I'm not sure your idea was to renormalize all the data from 2 different platforms into a unified set of expression values. If it was, I doubt that can work for a variety of reasons. Personally, I'd stay away from the "present/absent" concept, in any case. Do the studies you would like to reanalyze share a common control of some sort? I have published gene expression meta-analyses using similarities in experimental conditions as a way to compare expression profiles done on different platforms.

First of all THANK YOU EVERYONE!

You have helped a lot.

I would like to take account in consideration Micheaels warning, that was the reason why I thought about the simplier present-abscent method.

I have data set of A and B category in one study and data set from category A and C in the other, so I think I will do a metanalysis as A will be the control. Thank you again for help.