Deep Learning for Branch Point Selection in RNA Splicing

Branch point selection is a key step in RNA splicing, yet many many popular splicing analysis tools do not model this mechanism. There were relatively few confirmed branch points until 2015, when a genome-wide map of experimental human branch points was released. This data facilitates, for the first time, modeling branch sites with more sophisticated methods. We used deep learning to model branch site selection, which improved significantly over position-weight matrix models. We show that our branch point model can be used to classify potential disease-causing variants, and can help to improve existing splicing models.