Title: Analyzing and Predicting Alternative Splicing With Convolutional Neural Networks
Advisors: Georg Seelig and Larry Ruzzo
Abstract: Alternative splicing is a biological process by which a single mRNA transcript can be spliced into any of several forms, coding for proteins that can have different functions. This process is regulated by many factors, and its disregulation has been implicated in a number of diseases.
We focus on two main problems: discovering sequence motifs that affect alternative splicing, and predicting the effects of mutations on alternative splicing in vivo. Using a synthetic dataset, we trained a variety of neural network models in order to predict the effects of the exonic DNA sequence on splicing. Then, these models were used to derive sequence motifs that contribute to alternative splicing regulation across different cell types. For the prediction step, we tested the predictive power of these models as applied to a dataset of genomic mutations, and compared them with previous models. We will also discuss the construction of a database for alternative splicing events in the genome.