Item

Abstract

As a fruit of the current revolution in sequencing technology, transcriptomes can now be analysed at an unprecedented level of detail. But especially for the analysis of alternative splicing, there is still a lack of statistically robust methods to detect differentially spliced genes in RNA-Seq experiments.
In this work, we present two novel statistical tests to address this important methodological gap: a ‘gene-structure-sensitive’ Negative-Binomial (NB) test that can be used to detect differential transcript expression when the gene structure is known and a non-parametric kernel-based test, called Maximum Mean Discrepancy (MMD), for cases when the gene structure is is incomplete or unknown. Both methods also can cope with multiple replicates and can account for biological variance. Furthermore they can efficiently use paired-end read information.
We analysed both proposed methods on simulated read data, as well as on factual reads generated by the Illumina Genome Analyzer for two A.thaliana samples. Our analysis shows that the NB-test identifies genes with differential transcript expression considerably better than approaches based on transcript quantification, such as rQuant and Cuffdiff. But even more surprisingly we found that the MMD test performs as well as existing methods, in the absence of any knowledge of the annotated transcripts. This method is therefore very well suited to analyse RNA-Seq experiments, where other approaches fail, namely when the genome annotations are incomplete, false or entirely missing.
The software is available as a Galaxy package and as a standalone version.