Inference of Alternative Splicing

While traditional methods for computational recognition of alternative splicing are usually solely based on expressed sequences (ESTs or cDNAs; cf. Gupta et al., 2004, and references therein), more recent techniques try to identify and exploit local sequence features for prediction (Sorek & Ast 2003; Sakai & Maruyama, 2004; Dror, Sorek & Shamir, 2004; Sorek et al., 2004; Hiller et al., 2004). For instance, in Dror, Sorek & Shamir (2004) features like the exon length, its divisibility by three, the length of the flanking introns and the intensity of the poly-pyrimidine tract were utilized. Moreover, conservation patterns to another organism have been taken into account. Those are among the most discriminative features (Sorek & Ast, 2003). However, this is only possible for a fraction of exons in human, as exons are frequently not conserved, making the conservational features unavailable. In our previous work we aimed to design a classifier that accurately distinguishes constitutive from alternatively spliced exons. Our method only used information that is always available and might also be used by the cellular splicing machinery; namely, features derived from the exon and intron lengths and features based on the pre-mRNA sequence (Rätsch, Sonnenburg & Schölkopf, 2005).

We have now extended our work to other alternative splicing events such as intron retention, alternative 5’ and 3’ (intron) splicing. Furthermore, previously we only considered the model organism C. elegans. Now we have included higher organisms such as the zebra-fish, the plant Arabidopsis thaliana, and the fruit-fly. In this work we show that our method can accurately predict the mentioned types of alternative splicing for the four model organisms. We use our method to complete the inventory of known alternative splicing events with computational predictions.