Can I Get Geo Series Of Interests By Grouping Samples Together ?

Question

[Cassie]

Hi

I am wondering if it is possible to form a GEO series of my interest by grouping related GEO samples together and the resulted metrix file is comparable. For example, assume I am interested in healing rate of lung cancer pateints after chemo. If I pick some specific platform such as GPL571 and found all lung cancer GEO samples related to the platform, can I compare gene expression among these samples ? Of course, I have to pick ones which go through the same pre-processing step. Does it make any sense ? Or it is still uncomparable ?

Thank you very much,

Answer 1

[Sean Davis]

You might find the GEOmetadb bioconductor package to be useful for gathering samples of interest. If you are an R/Bioconductor user, the GEOquery package might be useful as well.

The data may be comparable, but it will depend on your question and the availability of samples within different GEO Series. For example, if you have tumor samples and normal samples, but they are from different GEO series, looking for differentially-expressed genes between tumor and normal is confounded by the effects of the experiments being done in different labs. Some of the genes might be truly differentially expressed, but some will simply appear so because of technical differences. If you want to go ahead, then you should absolutely get the raw .CEL files and start from scratch with quality assessment and normalization.

Another route that you could consider is to do a meta-analysis. Whether or not this will be possible will depend on the availability of suitable studies that meet the criteria for a meta-analysis that will answer your question.

Answer 2

[Neilfws]

What you're describing is meta-analysis: analysis of data from different, but related experiments.

There are lots of publications in this area and some useful tools in the Bioconductor suite. These include:

• GeneMeta
• metaArray
• RankProd

There is also MADAM in the CRAN repository.