Abstract

Alternative splicing is a mechanism that amplifies the library of 22,000 human genes to up to 1 million transcripts, most of which are regulated in a cell type-dependent fashion. Alternative splicing is known to control phenotypes such as sexual preference and neural connectivity and many disease mutations occur in splicing regulatory sequence. A central goal in the past decade has been to elucidate a ‘splicing code’: A set of RNA features in unspliced transcripts and a function that maps those features to tissue-dependent splicing patterns.

My group at the University of Toronto recently introduced a method that constructs a splicing code by extracting biologically informative signals from tissue profiling data and identifying many diverse RNA features that in combination are maximally predictive of those signals. Using the estimated inclusion levels of thousands of exons in diverse tissues, we inferred a splicing code from 987 putative RNA features. The utility and accuracy of the code is supported by experimental validation of predictions, plus feature evaluation using mutated minigene reporters. Our code includes previously-studied and novel features. Regulatory sequence identified by our method is located deeper into intronic sequence than previously expected and the number of regulatory features identified per regulated exon is surprisingly large. We believe that this code constitutes a first step in understanding the complexity introduced by ubiquitous alternative splicing.