Background: A new microarray platform (GeneChip Exon 1.0 ST) has recently been developed by Affymetrix http://www.affymetrix.com. This microarray platform changes the conventional view of transcript analysis since it allows the evaluation of the expression level of a transcript by querying each exon component. The Exon 1.0 ST platform does however raise some issues regarding the approaches to be used in identifying genome-wide alternative splicing events (ASEs). In this study an exon-level data analysis workflow is dissected in order to detect limit and strength of each step, thus modifying the overall workflow and thereby optimizing the detection of ASEs.
Results: This study was carried out using a semi-synthetic exon-skipping benchmark experiment embedding a total of 268 exon skipping events. Our results point out that summarization methods (RMA, PLIER) do not affect the efficacy of statistical tools in detecting ASEs. However, data pre-filtering is mandatory if the detected number of false ASEs are to be reduced. MiDAS and Rank Product methods efficiently detect true ASEs but they suffer from the lack of multiple test error correction. The intersection of MiDAS and Rank Product results efficiently moderates the detection of false ASEs.
Conclusion: To optimize the detection of ASEs we propose the following workflow: i) data pre-filtering, ii) statistical selection of ASEs using both MiDAS and Rank Product, iii) intersection of results derived from the two statistical analyses in order to moderate family-wise errors (FWER).