Introduction

The purpose of normalisation is to remove any of the non-biological, systematic effects arising in a microarray experiment, i.e. human, and technological errors. There are two types of error that methods proposed so far aim to correct: Spatial bias is mainly technologically related. Small errors that occur in the microarray robot, (e.g unevenly washing of the chip, inserting the chip at a slight angle in the scanner etc.), may have a large impact on the results. Many hours of printing might also alter the shape and size of the print-tips, and thus cause spatial patterns on the array.

Dye bias appear from the fact that the two dyes used, Cy$ 3$ and Cy$ 5$, have slightly different physical properties, and the react slightly different to photo-bleaching, an effect that occurs in multiple scannings of the array, [1]. Environmental factors such as the level of ozone, is also a contributing factor.

According to [1], a rule of thumb during normalisation should be to normalise all local features first, and then gradually progress to normalisation that involves several arrays. It is also recommended to correct for spatial effects before normalising for dye effects, since the dye and spatial effects can be confounded [1, page 133]. In SciCraft you can use the two nodes NormWithinArrays and NormBetweenArrays. The first concentrates on normalising within the arrays (local features), and the second one normalises between arrays (global features).

Bjørn Kåre Alsberg 2006-04-06