Machine learning annotation of human branchpoints

Abstract

Motivation The branchpoint element is required for the first lariat-forming reaction in splicing. However current catalogues of human branchpoints remain incomplete due to the difficulty in experimentally identifying these splicing elements. To address this limitation, we have developed a machine-learning algorithm - branchpointer - to identify branchpoint elements solely from gene annotations and genomic sequence. Results Using branchpointer, we annotate branchpoint elements in 85% of human gene introns with sensitivity (61.8%) and specificity (97.8%). In addition to annotation, branchpointer can evaluate the impact of SNPs on branchpoint architecture to inform functional interpretation of genetic variants. Branchpointer identifies all published deleterious branchpoint mutations annotated in clinical variant databases, and finds thousands of additional clinical and common genetic variants with similar predicted effects. This genome-wide annotation of branchpoints provides a reference for the genetic analysis of splicing, and the interpretation of noncoding variation. Availability and implementation Branchpointer is written and implemented in the statistical programming language R and is freely available under a BSD license as a package through Bioconductor. Contact b.signal@garvan.org.au or t.mercer@garvan.org Supplementary informationSupplementary dataare available at Bioinformatics online.