Poster Sessions

BIOINFO-17

Jennifer Barb

J. Barb, P. Munson

Elucidating Splice Variants Using the Affymetrix Human Exon Array: Effective Probeset Filtering Is Essential

Alternative splicing (AS) of mRNA is a major source of diversity of gene products in mammalian genomes, and is a part of a cell\'s programmed response to its environment in development or disease. The Affymetrix Human Exon 1.0 ST Array measures expression of ~106 individual exons, allowing screening for hundreds of potential splice variants. Data analysis for this platform is complex and problematic because of certain data artifacts. The conventional ANOVA model for Exon Array data includes factors for: Treatment effect; Variation within treatment group; Exon-specific probeset effect; and Treatment-Exon interaction. For the exons contained in each gene, this three-factor, mixed effect model is fit to transformed data and the significance and magnitude of the treatment-exon interaction is interpreted as evidence for AS. We have identified several shortcomings of this approach. Annotations (groupings of exons into genes) may be incorrect, grouping several distinct genes into one. Exons themselves may be incorrectly annotated. Low-intensity (\"dead\") probesets may not hybridize with their intended target. Surprisingly, many high-intensity probesets may also be “unresponsive” to treatment, possibly due to cross-hybridization. Each of these may contribute to false identification of AS. We assessed the severity of each problem with the following steps. The Exon array was re-annotated, based on conservatively defined RefSeq genes database. \"Non-core\" exons were excluded. Low-intensity probesets were identified and removed. Unresponsive probesets showing low variation across a wide variety of conditions in several datasets were excluded. In addition, we compared the AS gene lists identified under different filtering criteria to genes with known isoforms in public annotation sources as a means to validate our methods. We illustrate the benefits of this approach in screening for AS in an anatomical tissue dataset and THP1 cells incubated in the presence and absence of LPS (lipopolysaccharide).