InSilico DB has a "merging" R-Bioconductor package to combine public datasets from GEO. If you are not using R you can also combine data from the online platform (See this short step-by-step tutorial)

Example:

# Retrieve 2 datasets
eset1 = getDataset(gse="GSE10072", gpl="GPL96", norm="ORIGINAL", genes=TRUE);
eset2 = getDataset(gse="GSE7670", gpl="GPL96", norm="ORIGINAL", genes=TRUE);

#combine them
esets = list(eset1, eset2);
eset = merge(esets, method="NONE");

#plot them
plotMDS(eset, targetAnnot="Disease", batchAnnot="Study");

InSilico DB packaged various batch removal effects methods so line 4 could be replaced with:

eset = merge(esets, method="XPN");

# or

eset = merge(esets, method="COMBAT");

Hope this helps.

For more info Bioinformatics paper reference; InSilico DB and InSIlico Merging packages links, and blog link.

• Unlocking the potential of publicly available microarray data using inSilicoDb and inSilicoMerging R/Bioconductor packages - BMC Bioinfomatics
• inSilicoDb: an R/Bioconductor package for accessing human Affymetrix expert-curated datasets from GEO - Bioinformatics

-Tutorial example : https://insilicodb.org/the-impact-of-batch-effects-when-merging-different-data-sets/

R-Bioconductor packages:

• http://www.bioconductor.org/packages/2.12/bioc/html/inSilicoDb.html
• http://www.bioconductor.org/packages/2.12/bioc/html/inSilicoMerging.html

You can certainly download all the .CEL files and normalize them together. However, you may find that your hypothesis testing could be challenging since there will likely be a batch effect between the two datasets.

In theory if you have two sets of raw expression sets, from the same array model, then you can simply bind one to the other (accounting for the probe location). However, doing this creates a whole world of problems. There would have to be a very good justification for doing this. The first problem is that there will be a batch effect between the two datasets, as previously mentioned. If you manage to correct for that successfully then, you might get some meaningful data out of the analysis, might. This is a post experimental design decision and is not recommended.