Unbiased characterization of high-density oligonucleotide microarrays using probe-level statistics.

Affymetrix GeneChips are being used increasingly for quantitative monitoring of gene expression in a variety of biological systems. Depending on the experiment, the analysis of Affymetrix results can have several different goals ranging from calculation of signal strength for a variety of inter-gene comparisons to the determination of which genes show significant differential expression between sample conditions. There have been several proposed methods for precise quantification of expression signal with promising results; however the question of what constitutes a significant change between replicate groups still remains. We have designed a method which performs statistical analysis on the differential expression of genes in the Affymetrix GeneChip system at the probe level in order to bypass the assumptions made in other analysis techniques. Validation using both spike-in data and real experimental data proves the method is effective at isolating differentially expressed genes statistically, thereby eliminating the need for arbitrary restrictions such as fold change. Application to an existing neural stem cell data set demonstrates the method's applicability to highly complex systems and its ability to detect very low expression differences (<1.2-fold change), providing resolution which may be of significant interest in neural systems such as this.