Prediction of alternatively spliced exons using Support Vector Machines

Abstract

Alternative splicing is a mechanism for generating different gene transcripts (called isoforms) from the same genomic sequence. In recent years, it has become obvious that a large fraction of genes undergoes alternative splicing. Finding alternative splicing events experimentally is both expensive and time consuming. Furthermore, traditional transcript-to-genome alignment-based approaches are limited to complete and annotated genomes for which a large amount of transcript data is also available. As a complement to experimental and traditional computational approaches, newer machine learning approaches have been successfully applied to the problem of predicting alternative splicing events. In particular, previous work has shown that machine learning algorithms that utilise sequence features can be used to learn classifiers for distinguishing alternatively spliced exons from AQ1 constitutively spliced exons. In this paper, we explore the predictive power of a large set of gene features that have been experimentally shown to have effect on alternative splicing. We use such features to build Support Vector Machine (SVM) classifiers for predicting alternatively spliced exons and constitutive exons. Experimental results show that classifiers built from the extended set of features give better results than those that consider only Basic Sequence Features (BSF). Furthermore, we use feature selection methods to identify the most informative features for the prediction problem at hand.

Kassiopeia: a database and web application for the analysis of mutually exclusive exomes of eukaryotes

“…In contrast to these high-throughput experimental data methods, computer based de novo predictions of alternative splicing events are not well established yet. In one approach support vector machine classifiers have been built from gene features that have experimentally been shown to effect alternative splicing [11]. Other approaches used bayesian networks to predict NAGNAG tandem acceptor splice sites [12], genetic programming to classify cassette exons versus retained introns [13], and ab initio gene prediction methods [14].…”

Genome-wide analysis of mutually exclusive splicing

“…In contrast to these high-throughput experimental data methods, computer based de novo predictions of alternative splicing events are rather complicated. In one approach support vector machine classifiers have been built from gene features that have been experimentally shown to effect alternative splicing [208]. Others used bayesian networks to predict NAGNAG tandem acceptor splice sites [209], genetic programming to classify cassette exons versus retained introns [210], and ab initio gene prediction methods [211].…”