Conclusion
Single isoform gene structure input improved precision of ExAlt predictions. ExAlt reported overlapping false positive exons in only 4% of the constitutive exons (87% of the remaining exons were correctly classified as constitutive exons), while identifying nearly two thirds of the overlapping alternatively spliced exons. With limited gene structure information, ExAlt detected overlapping exons with modest increases in false positive predictions. Limiting over prediction in a non-expression based alternative exon finder proved to be a challenge since multiple overlapping sequence intervals are potentially predicted as exons. The problem is illustrated by the performance of the ab initio versions of ExAlt, which correctly predicted overlapping exons, but at the expense of increased over prediction of splice sites. Supplementing the statistical sequence models with evidence from cross-species sequence conservation proved to be an effective strategy in reducing the number of false positive predictions, while maintaining sensitivity levels.
In model species such as Drosophila melanogaster, significant effort has gone into providing accurate gene structure annotations, which account for many of the proteins present in the cell. With the abundance of alternative splicing already known to occur, it is likely that evidence for new examples of alternative splicing will continue to grow. Therefore, even with the availability of high quality annotations, there are new variations in gene structure yet to be discovered and non-expression based prediction methods such as ExAlt can be used to search for new cases of alternative splicing.