Normalization of microarray expression data in cDNA data & how procedure is performed

[ManGer]

What is meant by the ‘normalization’ of microarray expression data? Illustrate your answer with some examples of how this procedure is performed for the case of dual dye (2 color) cDNA microarray data.

Thanks for any help.

[tim007]

Normalization means to adjust microarray data for effects which arise from variation in the technology rather than from biological differences between the RNA samples or between the printed probes.

Composite Loess Normalization
Diagram shows the loess curve through a series MSP titration spots. Yang et al's paper propose the normalization:

Where N = M – p(A) loessMSP(A) – {1 – p(A)} loessi(A)

Where loessMSP(A) is the loess curve through the MSP spots and p(A) is the proportion of spots on the array with A-values less than A. Here, normalization will be increasingly based on the global MSP curve rather than the individual tip-group curves with the smaller number of spots. Composite normalization procedures use constant local regressions (degree=0) to construct the MSP loess curve so that any necessary extrapolation of the MSP curve will also be constant


Housekeeping Gene Normalization
One can calculate normalization coefficient using the average of the housekeeping genes fluorescence intensities:

k = mE
mC

Where:
mE: mean (average) of the selected housekeeping genes
mc: mean of the selected housekeeping genes for the control.

Ideally, you should use the normalization software that was provided by the vendor of the array scanner. If you used Affymetrix chips and have the CEL-files with the measured intensities, you can also simply compress these files in a zip-archive (using a freeware compression software like WinZip) and upload them on e.g. ArrayMining.net and the software will automatically apply the Robust Multiarray (RMA) normalization method. Other normalization methods for two-color arrays include Gene Chip RMA (GC-RMA) and Variance Stabilising Normalization (VSN), for example, however there is no agreement in the scientific community on a single "method of choice". Alternatively, you can apply multiple standard approaches, like RMA, VSN, MAS5, etc. separately and compare the results.

If you would like to analyze the normalized data with e.g. ArrayMining.net, it is also important to note that in particular for the supervised analysis modules (Gene Selection, Class Assignment Analysis) you need multiple replicates per biological condition (because the algorithms need to estimate the within-class-variance in expression values). To identify two-fold differences in gene expression across biological conditions, I recommend to use at least 3 microarray samples per biological condition (without pooling).

P.S. I like this part of the experiment having mastered it all the time!