Tools and approaches for an end-to-end expression array analysis

Abstract

Microarray experiments can appear daunting because the considerations called for in their analysis cover several fields of research. To understand the data microarrays generate some knowledge of classical statistics and recent complexity theory are useful while emerging computational techniques such as XML directed workflows could aid in managing the data. These considerations are called for because as experimental tools, microarrays (arrays) exemplify the recent trend in biological research towards high dimensionality datasets. Until recently observations were made on only a few variables at a time and these were used to support or refute hypotheses, but high dimensionality datasets are generated by observing a very large number of variables (e.g. gene expression measurements) at the same time. The number of expression measurements made on arrays is not only high, but notably high when compared to the size of a typical sample population. This combination of high dimensionality and asymmetry leads to large datasets and fundamental problems when using standard approaches to interpret the data. An end-to-end approach is a general framework in which to place some useful considerations when planning an analysis. The framework described here explores the origins of signal and several sources of variance, approaches to representing high-throughput data, the statistical considerations when modeling array data and the software tools that can aid in carrying out the analysis.