GenePC AND ASPIC INTEGRATE GENE PREDICTIONS WITH EXPRESSED SEQUENCE ALIGNMENTS TO PREDICT ALTERNATIVE TRANSCRIPTS

We have developed a generic framework for combining introns from genomicly aligned expressed–sequence–tag clusters with a set of exon predictions to produce alternative transcript predictions. Our current implementation uses ASPIC to generate alternative transcripts from EST mappings. Introns from ASPIC and a set of gene predictions from many diverse gene prediction programs are given to the gene prediction combiner GenePC which then generates alternative consensus splice forms. We evaluated our method on the ENCODE regions of the human genome. In general we see a marked improvement in transcript-level sensitivity due to the fact that more than one transcript per gene may now be predicted. GenePC, which alone is highly specific at the transcript level, balances the lower specificity of ASPIC.