Normalisation methods

The normalisation process aims at balancing many of the systematic variations present in an array experiment (Quackenbush, 2002).
In recent years, several normalisation techniques have been developed (Kroll and Wolfl, 2002).
The underlying principle behind the different methods requires the identification of genes that are not affected by experimental conditions and, thus, show a ratio equal to 1 between the reference and the experimental sample. To compare measurements from gene expression array experiments, quantitative data can be normalised using reference genes or global normalisation methods based on mean or median values. These methods are based on the assumption that:

1. selected reference genes are expressed at a standard level in all experiments or 
2. that mean or median signal of expression will give a quantitative reference for each individual experiment (Quackenbush, 2001).

In addition to these methods, there are a number of alternative approaches including:

• linear regression analysis
• log centering
• rank invariant methods
• Chen’s ratio statistics
  

However, none of these approaches takes into account systematic biases that may appear in the data. Several reports have indicated that the log2(ratio) values can have a systematic dependence on intensity. Lowess normalisation is a locally linear robust scatter plot smoother generally preferred to an overall linear correction because it compensates for variation of the correction factor with signal intensity and is largely unaffected by outliers, including the regulated transcripts. 

Normalisation references

• housekeeping genes

Genes whose expression is expected to be constant across samples adopted as internal controls. However, recent studies have demonstrated that genes that do not change in their expression levels in response to a variety of experimental conditions simply do not exist. Another approach to control variability in microarray experiments is to spot genomic DNA in multiple dilutions on the array and use the signal obtained after hybridisation for normalisation purposes (Benes and Muckenthaler, 2003).

• exogenous control genes

External controls can be used as negative controls if no corresponding mRNA is present in the RNA samples to be analysed; negative controls help to determine the noise of a microarray experiment. In some cases

• internal controls

Positive or “spike-in” controls: exogenous RNAs added to the reference and the experimental samples in predetermined concentrations before the synthesis of fluorescent-labelled cDNAs. Spike-in controls need to titrated to cover the entire range of signal intensities obtained in a microarray experiment to be representative for all detectable genes. Spike-in RNAs added in equal amounts to the reference and experimental sample can serve as normalisation controls (Benes and Muckenthaler, 2003).