using of Rankprod to combine two different data sets with different affy platform numbers

Question

[rajasekargutha]

can i use Rankprod package to combine two different data sets with different affy platform numbers

Answer 1

[Kevin Blighe]

'Combine' is the incorrect word here. RankProd compares lists of terms by the rank of their statistical significance. It does not combine expression values into a merged dataset, if that's what you want.

Take a look at Section 3.2 of the RankProd manual, where it states:

Data with multiple origins

In this section, we will illustrate how the RP method can be applied to datasets containing samples from multiple origins using the built-in data set arab. As mentioned before, arab consists of array data measured by two different laboratories. Both laboratories measured gene expression under the same two conditions.

Given the lack of experimental standards for microarray experiments, direct comparison is not feasible. Instead of using actual expression data, our approach combines the gene rank from different origins together (for details refer to Breitling et al. (2004)).

Thus, you can compare the results from your two Affymetrix microarray experiments. You may have to filter the datasets to only include common probe IDs prior to running RankProd, but you'll find out when you run it. It may also be advisable to summarise expression values over genes and/or transcript isoforms, and converting probe IDs to HGNC gene names, prior to running RankProd in this way.

[rajasekargutha]

Call I use data from GPL96 and GPL 570 platform as two origins data.

[Kevin Blighe]

As they are both Affymetrix Human Genome U133 arrays, then I don't see a major problem. Many of the probes will be identical on both. I do recommend filtering to just keep common probes though. You can use the Affymetrix probe IDs to see which are common and which are not

If in further doubt, contact the developer of RankProd

[rajasekargutha]

Could you explain how to remove probes that are not present in two datasets

[Kevin Blighe]

When you normalise each dataset, you will have an expression set object. The rownames of these will typically be Affymetrix probe IDs.

You can check which probes are common via the which or match functions in R. Be careful, though, as these functions function very differently.