Microarray RMA_log2Signal fold change calculation

Question

[Jaccy]

Hi dear biostars.

I got hand on some microarray data which is already "user-friendly" processed. Unfortunatelly, I get very nervous if I'm supposed to work with data I don't understand. I will try to explain, what I think I understood so far.

So I have some columns with "RMA_log2Signal", which is already normalized. Then there is another column with "fold change".

When I calculate

1. Signal control - Signal experiment = signal fold change (e.g. 1,6-11,3 = -9,7)
2. 1/(2^signal fold change) (1/2^(-9,7) = 832)

I get the "fold change" documented. Does this mean that my gene of interest is 832 fold higher express than in the control condition?

Thanks for your advice, or also google tipps...

Cheers, Jaccy

[Kevin Blighe]

Hello Jaccy. Which array type and version is it? Mainly interested to hear if it is single- or two-channel (colour).

[Jaccy]

Hey Kevin, it's single-channel (Affymetrix Gene chip)

[Kevin Blighe]

Cool - great. The advice that Ahill gives is pretty good. I should say, too, that you could technically keep just the RMA_log2Signal columns and still proceed to using, for example, limma, for the purposes of doing a differential expression analysis. Limma will calculate moderated statistics via empirical Bayes. Currently, it is obvious that the fold change that is in your data was calculated manually.

You would obviously also only keep these 'RMA' columns when generating box-and-whisker plots, histograms, cluster dendrograms, etc.

Also remember that the 832 is a linear fold change difference - pretty high but can happen with this type of data.

[Jaccy]

Thank you so much for the advices, this will help me a lot!

Answer 1

[Ahill]

Yes, given the types you describe. RMA is on the log-2 scale. You've correctly calculated the fold change. 832 is a very big fold change, so might be very surprising but perhaps not impossible, depending on what your gene is and what the experimental conditions are. I'd agree that it is risky to use data if you don't have the detailed methods that were used to produce it. To increase your comfort level or reveal issues with the data, if you haven't already done it I'd strongly recommend making plots and summary statistics to confirm the data is well normalized. Some useful ones to start would include: boxplot the RMA_log2Signal values across all samples, make M-A plots for any comparisons of interest. For google, a query like "quality control plots for microarray gene expression data" would turn up some good advice as well; or take a look at R packages like arrayQualityMetrics or others.

[Jaccy]

Ahill, thank you so much for your answer!

I think I will start with your tipp to google "quality control plots for microarray gene expression data" and then jump deep into the data myself. This will make me feel better in any case

Yeah, the high expression made me suspicious, too... However, I work with plants and they suffered a lot, nevertheless, I will check again

Thank you for your help, Ahill!

[Kevin Blighe]

I have moved Ahill's comment to an answer. Please feel free to up-vote and/or accept the answer. Best of luck!