A framework for the informed normalization of printed microarrays

Microarray technology has become an essential part of contemporary molecular biological research. An aspect central to any microarray experiment is that of normalization, a form of data processing directed at removing technical noise while preserving biological meaning, thereby allowing for more accurate interpretations of data. The statistics underlying many normalization methods can appear overwhelming to microarray newcomers, a situation which is further compounded by a lack of accessible, non-statistical descriptions of common approaches to normalization. Normalization strategies significantly affect the analytical outcome of a microarray experiment, and consequently it is important that the statistical assumptions underlying normalization algorithms are understood and met before researchers embark upon the processing of raw microarray data. Many of these assumptions pertain only to whole-genome arrays, and are not valid for custom or directed microarrays. A thorough diagnostic evaluation of the nature and extent to which technical noise affects individual arrays is paramount to the success of any chosen normalization strategy. Here we suggest an approach to normalization based on extensive stepwise exploration and diagnostic assessment of data prior to, and after, normalization. Common data visualization and diagnostic approaches are highlighted, followed by descriptions of popular normalization methods, and the underlying assumptions they are based on, within the context of removing general technical artefacts associated with microarray data.