Bayesian De Novo Characterization of Alternative Splicing Using High-Throughput Sequencing

• Activity Number: 347
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): David Rossell*+ and Camille Stephan Otto-Attolini and Manuel Kroiss and Almond Stöcker
• Companies: IRB Barcelona and IRB Barcelona and LMU Munich/TU Munich and LMU Munich
• Keywords: model selection ; alternative splicing ; high-throughput sequencing

Abstract

High-throughput sequencing offers an unprecedented resolution to study gene alternative splicing (AS). AS has important implications for human health, e.g. it is involved in multiple diseases and cellular malfunctions. De novo characterization of AS, i.e. not conditioning on a set of known variants, poses very serious statistical challenges. The main issue is that the problem, which can be seen as a particular case of model selection, has a supra-exponential number of possible models.

Our proposed Bayesian framework improves upon previous approaches by (i) strongly enforcing parsimony by specifying non-local priors (NLPs) and (ii) incorporating valuable, publicly available, prior knowledge. NLPs have been shown in previous work to have good theoretical properties for sparse model selection. Integrating the experimental data with the accumulated prior knowledge allows to recover the set of expressed variants at a lower sequencing depth. We explore large posterior probability models via a random-walk MCMC tailor-suited to this problem. The framework is computationally efficient and shows good performance.