No, normalization is not the same, and other data processing steps also are usually different (e.g. background correction).

I strongly recommend using the raw data and use a common data processing pipeline to combine different data sets.

Also be aware that there are different platforms using different amplification and detection chemistries, and they might not be comparable at all. You must check that the protocols are comparable. This may not be a severe problem if you have 100's or 1000's experiments you are going to analyze - in this case you must check that the methods/protocols are not confounded with conditions/groups you are analysing .

If you only use fold-changes (a simple meta-analysis, so to say), the problems are less severe since at least some of the method-specific impact cancel out in the"within-experiment" fold-changes. But here I still recommend having a rather large and diverse data set (100's of experiments or so), and I still would check for confounding issues.

Whatever you do... it will be a tough job without knowing R at all. You can surely do "something", come up with some kind of pretty heatmap or whatever and draw interesting conclusions, but this will likely just be a part of "irreproducible research". It might be a good idea to cooperate with an experienced biostatistician.

Daphne Olivia Jawai Sadai

The normalization is not same in microarrays. The different normalization methods are available for microarrays (for example RMA, GCRMA, MAS5, quantile). You can find a lot of resources from the web for microarray data analysis. The R packages (for example limma, oligo) are used for normalization and analysis.

Hello Daphne,

if you are going to use GEO2R you have always the same normalization, implemented from GEO. From normalization view this is comparable, but you should be aware, that the selected datasets are comparable. In GEO there are different chip types, they are differently hybridized and do not always include data of the same tissues. This is much more interesting than normalization. So choose carefully the right GEO data, use GEO2R and get comparable outputs of significant gene lists. Also be aware to analyse only two chip groups for one analysis: You could select this in GEO2R. Nevertheless these lists could differ extremely, because hybridized in different laboratories.

Not satisfied with GEO2R or not able to manage it, you could download the CEL files from GEO and achieve quantile normalized expression values with RMAExpress (http://rmaexpress.bmbolstad.com, an easy to install Windows tool) of all chips and all genes uploading these normalized Signals to webtool ArrayMining (arraymining.net; use Gene selection). Upload the complete Signal dataset for one analysis (2 groups), choose an analysis method and you will get a quality ranked list of significant genes. The best combinations could be found in our analysis on ResearchGate: https://www.researchgate.net/publication/276981219; I would recommend eBayes or SAM. As you could see in this publication combining different normalizations and gene expression profiling methods could differ extremely in some cases (down to 7 of 999 identical genes, in average 26.5% identical genes in two analyses). Therefore, if you found different datasets on GEO use always the same methods.