Genome-wide detection of alternative splicing in expressed sequences using partial order multiple sequence alignment graphs.

We present a method for high-throughput alternative splicing detection in expressed sequence data. This method effectively copes with many of the problems inherent in making inferences about splicing and alternative splicing on the basis of EST sequences, which in addition to being fragmentary and full of sequencing errors, may also be chimeric, misoriented, or contaminated with genomic sequence. Our method, which relies both on the Partial Order Alignment (POA) program for constructing multiple sequence alignments, and its Heaviest Bundling function for generating consensus sequences, accounts for the real complexity of expressed sequence data by building and analyzing a single multiple sequence alignment containing all of the expressed sequences in a particular cluster aligned to genomic sequence. We illustrate application of this method to human UniGene Cluster Hs.1162, which contains expressed sequences from the human HLA-DMB gene. We have used this method to generate databases, published elsewhere, of splices and alternative splicing relationships for the human, mouse and rat genomes. We present statistics from these calculations, as well as the CPU time for running our method on expressed sequence clusters of varying size, to verify that it truly scales to complete genomes.