# Duplicate gene symbol handling in GEO gene expression data

I have downloaded a gene expression data from GEO database (GSE3268) in which in some of its rows there are duplicate gene symbols.

For example TP53 exists in two rows with different expression values and different GenBank Accession IDs.

They have the following Nucleotide Title:

Human p53 cellular tumor antigen mRNA, complete cds

Homo sapiens tumor protein p53 (TP53), transcript variant 1, mRNA

I want to map these data with edge list of gene symbols to make a network. Which expression value should I consider for such duplicate gene symbols for mapping?

## 1 Answer

Since this is a microarray (U133A), what you're seeing is that each probe is associated with one or more transcripts. The general strategy when dealing with microarrays is to use RMA (or fRMA) to summarize to probesets and then collapseRows to summarize to gene-level information. Since this is all typically done in R, you can find some further discussion on the bioconductor site. I would suggest going through this process and then only using the result of that in building your network (whether you're happy with sticking to transcript-level quantifications or gene-level quantifications for building your network is up to you).

• I have a problem with using fRMA since the there is no supplementary .cel file in GEO database for this dataset. How can I use R on .SOFT files to summarize gene information? Another point is that in one of the subsets of my data set, (ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM73388), it seems that RMA have been used and data is collected in .SOFT file, but again there is duplicates in it. Dec 25 '17 at 19:13
• If RMA has already been used then you needn't run it a second time. Dec 27 '17 at 20:52
• Very useful answer! In EuropePMC I found 63 publications mentioning the use of collapseRows(). Many don't report the setting they apply. Some high impact publications reported as follows: "MaxMean" (default) = Gandal (Science2018), Raznahan (PNAS2018), Langfelder (NatNeurosci2016); "maxRowVariance" = Langfelder (NatNeurosci2016) Jan 15 '19 at 3:27