Is there a standard approach to normalizing across various platforms in NCBI's GEO? I was thinking I could just push them all out of GEO and into limma as expression objects, but I wasn't sure if someone had addressed this issue yet. I would obtain them as raw data to ensure uniform normalization.
This will be somewhat dependent on the platform (affymetrix, illumina, agilent, etc.) and also the type of experiment (expression microarray, methylation array, etc.). Having said that, it's likely that you mean affymetrix expression microarrays, in which case I would generally use GCRMA or RMA (it's in the affy package).
Just gene expression experiments. My goal is to show that our model is similar to models of related diseases than you'd expect by chance. My plan would be to pull down each experiment individually (there aren't many) and then push them into limma to do the same normalization method.
If I understand correctly, the best practice would be to normalize the arrays under the same method to make them directly comparable. I can't use affy because our study data comes from agilent single color arrays.
Should I normalize each study independently, or should they all be normalized together?
RMA, at least in an adapted form, can still be applied to at least some Agilent microarrays and seems to perform well. Aside from that, the normalizeBetweenArrays() function in limma can probably handle many of the normal use cases (it's best to make a number of diagnostic graphs so you can ensure that whatever normalization strategy you use does a good job...most of the vignettes on Bioconductor will provide examples of how to do this).
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
What format did you get them from from GEO? It's often the case that they've already been normalized if you get a series matrix file. in that case, be careful if you need to analyze multiple GEO datasets together, as it's not uncommon to find they've been normalized differently (in which case you need the raw data).
I've updated the OP. I would pull down the raw data so I can just do the normalization and avoid that issue.