I have a set of gene expression data downloaded from http://www.ncbi.nlm.nih.gov/geo. I have two sets of data, one is the raw probe intensity data set in the form of CEL files. Another is processed txt file.

Here is a link for example http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE7904

I don't know what kind of processing is done to generate the processed data from the raw CEL files. For example in the above link, I have processed data like

SOFT formatted family file(s) MINiML formatted family file(s) Series Matrix File(s)

I don't know what sort of processing is done to generate those files. Any suggestions? I can read the raw CEL files in matlab but what sort of preprocessing am I supposed to do?

I also want to map these probe its to gene ids and get the corresponding gene. How can I accomplish this? I know matlab. But I am bit confused about the terminology and all. Any suggestions?

  • $\begingroup$ this website converts affy id's to various gene id codes if this is the sort of thing you are after $\endgroup$ – rg255 Feb 11 '13 at 21:15
  • $\begingroup$ @rg255. I didn't get how the processed data is generated. As you can see here ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE7904 there are two data sets. One is raw CEL files and the other is Series Matrix File(s). How is this processed data generated. Also if I look at the number of rows. In the case of Cel files, the number of probes is huge 1354896. But when I look at the series matrix data it is around 50000. I didn't get it. Any suggestions? $\endgroup$ – user34790 Feb 11 '13 at 21:31
  • $\begingroup$ @rg255. Also I didn't get the link you said this website. I didn't find the link attached $\endgroup$ – user34790 Feb 11 '13 at 21:31
  • $\begingroup$ oops, david.abcc.ncifcrf.gov/conversion.jsp - as for the rest of your question I wouldn't like to answer as I am not an expert $\endgroup$ – rg255 Feb 12 '13 at 7:17

If you click through to some of the samples in the study, eg. GSM194397, GSM194398, etc. it mentions under the "Data processing" section that "The data were analyzed with dchip with default normalization settings".

You can learn more about dChip over at the dChip website.

  • $\begingroup$ Thanks I will try that. However it looks like I will need some CDF files as well. Where can I get them? In particular I need the CDF files for chip type HG-U133_Plus_2 and HG-U133A. Any suggestions? I looked a couple of places but looks like I have to pay for it. It was supposed to be free I guess $\endgroup$ – user34790 Feb 11 '13 at 22:01
  • $\begingroup$ I suggest you use the limma package in bioconductor to analyze the data. I'm not sure why one study has data from two chips -- it's unfortunate. You'd first have to study each batch separately, then do some meta-analysis to combine the results from each together. If you're not comfortable programming, you might try Gene Pattern, though I've never used it. Lastly: you should likely post bioinformatics related questions over on biostar. $\endgroup$ – Steve Lianoglou Feb 12 '13 at 2:36
  • $\begingroup$ I need to get the gene id for the corresponding probe it. For that I need the CDF file. So I was actually wondering where I can get them? $\endgroup$ – user34790 Feb 12 '13 at 5:53
  • $\begingroup$ There are other ways to convert id's to genes. I'd be surprised if you can't query BioMart to get that answer (look in the ID Converter tab). It also looks like the DAVID website has a conversion tool that may get you what you need. $\endgroup$ – Steve Lianoglou Feb 12 '13 at 16:37

The methods used for the processign of the date should be mentioned in the Methods section of the paper the data come from (here it is Richardson AL, Wang ZC, De Nicolo A, Lu X et al. X chromosomal abnormalities in basal-like human breast cancer. Cancer Cell 2006 Feb;9(2):121-32. PMID: 16473279). I quote from the paper:

"RNA extraction, cRNA synthesis, and hybridization to Affymetrix Human Genome U133 Plus 2.0 Arrays were performed as described previously (Signoretti et al., 2002 and Wang et al., 2004). Raw expression data obtained using Affymetrix GENECHIP software was normalized and analyzed using DNA-Chip Analyzer (dChip) custom software (W.H. Wong and C. Li, http://www.dChip.org/). Array probe data were normalized to the mean expression level of each probe across a sample set. Where indicated, tumors were classified as BLC or non-BLC on the basis of their expression array characteristics, using dChip hierarchical clustering analysis as previously described (Matros et al., 2005 and Wang et al., 2004). Comparisons between results obtained on BLC or BRCA1 tumors, non-BLC tumors, and normal breast samples were performed using the dChip “Compare Sample” function. A threshold of 1.2-fold overexpression in BLC and BRCA1 tumors was applied with 90% confidence. Of 1271 gene probes that map to the X chromosome, 60 satisfied these overexpression criteria with a range of fold difference from 1.35 to 5.11. The false discovery rate (number) of 1000 permutations was as follows: median, 0% (0); 90th percentile, 3.3% (2). Of the 60 probes, 19 were redundant (two or more probes mapping to the same gene) and excluded, leaving 41 gene-specific probes for use in the expression plot of Figure 5B. The complete gene expression array data set is available on the NCBI GEO database (accession no. GSE3744)."

Since this is a standard Affymetrix human genome chip, it's CDF should either be embedded in most packages used to analyse this kind of data (limma is an industry gold standard example ;-), or downloaded from the Afymetrix website. Please note that the CDF itself may be quite old and it's annotation not always accurate any more (I have no experience with human data), so it's worth using an updated CDF (see Dai et al 2005).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.