Singular Value Decomposition–Based Alternative Splicing Detection

Abstract

Altered alternative splicing has been identified as an important factor in tumorigenesis. The Affymetrix exon tiling array is designed for detecting alternative splicing events in a transcriptome-wide fashion; however, there are currently few analysis tools that have been well studied for effective detection of alternative splicing events. We propose a new screening procedure based on singular value decomposition (SVD) of the residual matrix from a robust additive model fit to probe selection region (PSR) data. With this approach, we analyze the exon tiling array data from a brain cancer study conducted at the M. D. Anderson Cancer Center, and show that the proposed SVD-based approach is able to better accommodate outlying measures and capitalize on the multidimensional group-by-PSR gene expression profiles for more effective detection of group-specific alternative splicing events, as well as the PSRs most likely associated with the alternative splicing. Lab validation confirmed some of our findings, but the list of candidates detected with our proposed method may provide a better signpost to guide further investigations.