RankProduct - Is this a valid analysis?

[Randy912]

Hello,

I analyzed a few GEO data sets that were on different microarray platforms using the RankProd R library.

However, since the data sets were on different microarray platforms, they have different probe sets and hence there were tons of NAs in the all-inclusive expression set. The candidates I ended up taking from this analysis actually ended up working, but I'm wondering if this analysis is defensible.

Thank you in advance for any insight

[IonTom]

From my experience Rank Product test gives much better and reliable results than most parametric methods.

To your question. You should limit both microarrays to the probes that are available on both platforms. Otherwise the number of probes will be different for both platforms and the ranks will not be comparable.

But limiting the probe sets should help. If the probes are nor directly matchable,
you could use the gene level expression for a comparison.

[Randy912]

The microarray platforms have totally different probe annotations, so there are actually 0 probe sets shared across all platforms.

When you say "use gene level expression," how do you change the probe set names to facilitate the matches? For example, let's say gene X has probes 38548_at in an affy array, and 294838 in an Illumina array, how do you modify them so that the comparison can be made?

Thank you for your help.