RMA with Affymetrix Expression Console vs Oligo package

Question

[User43533445]

I have 192 .CEL files (chip: Affymetrix Human Clariom S HT ). If I apply the RMA using Affymetrix Expression Console software (Analysis: Gene Level - SST-RMA) or with the rma() function from the Oligo package in R Bioconductor, I get quite different results. In general, the log2 expression values I obtained with Affymetrix Expression Console are higher than the values from the Oligo package, although the pattern of expression levels per transcript is the same. What is the reason for this difference and how should I decide which of the two methods to use?

Here's an example of the expression levels of a transcript for the first 7 arrays:

Oligo:

 TC0100006675.hg.1  9.96045562120938    10.0166037327827    10.2163252874566    10.0278834737392    9.79302430710618    9.74687422664869    9.86460893871565

Affymetrix Expression Console:

TC0100006675.hg.1   14.02148    14.2837    14.33652    14.2356    13.81137    13.6973    13.91579

With Oligo, I applied the rma() function on the entire data set, including all spike-ins/control probes. I'm not sure if that's the case in Affymetrix and if that's the correct way to do so.

Thank you for your reply!

[rodrigohb.93]

Hey, did you find the answer to your question?

[Hamtaro]

Did you find an answer. I'm having the same problem as you and I don't know how to fix this.

[Kevin Blighe]

The answer is below. Different normalisation methods are being utilised.

[Hamtaro]

Thank you Kevin. Could you help me with this question? What to do with the empty rows of results obtained with the Affymetrix Expression Console? Nobody replies me..

[Kevin Blighe]

It seems that you now have a reply? / Parece que ahora una persona te haya respondido?

[manaswwm]

For anyone who stumbles on this post in their quest to find an answer to this question - https://assets.thermofisher.com/TFS-Assets/LSG/brochures/sst_gccn_whitepaper.pdf

SST "decompresses" the probe intensities (read - increases intensities) to make it more comparable across different technologies. Since RMA does not do this, the signal seems to be under-represented when compared to SST-normalized. But, if I understand correctly, both given similar signals in terms of biology with a difference in the intensities

Answer 1

[Kevin Blighe]

There are a few possibilities; however, the glaring one is that your data is simply being processed differently between oligo and the Expression Console.

Signal Space Transformation - Robust Multiarray Average (SST-RMA) applies some pre-processing steps to your data before normalising via RMA. SST-RMA was introduced to address 'fold change compression', i.e., under-estimation of true fold-changes.

That explains the difference.

Kevin