Evaluation of the exon predictions of the GRAIL software

Identification of transcribed sequences within genomic regions has been a major rate-limiting step in the pursuit of genes. We have tested the ability of GRAIL 1 and GRAIL 2 to locate coding exons in the 41 longest human genomic sequences available in release 38 of the EMBL database. Of the total annotated exons GRAIL 1 correctly identified 68% of coding exons. GRAIL 2 correctly identified 71%. Of the 686 predictions made by GRAIL 1, 46.4% did not correspond to an annotated exon. The corresponding figures for GRAIL 2 were 556 and 30.5%. The GRAIL 2 software predicted as excellent a significantly higher proportion of the coding exons than did GRAIL 1, and also gave a significantly lower figure for false predictions. GRAIL 2 thus represents a significant improvement over GRAIL 1.