Personal tools
You are here: Home Events ANC Workshop: Yuanhua Huang and Guido Sanguinetti, Chair: Scott Lowe

ANC Workshop: Yuanhua Huang and Guido Sanguinetti, Chair: Scott Lowe

— filed under:

What
  • ANC Workshop Talk
When Dec 01, 2015
from 11:00 AM to 12:00 PM
Where IF 4.31/4.33
Add event to calendar vCal
iCal

Yuanhua Huang 

"Statistical modeling of isoform splicing dynamics from RNA-seq time series data"

In most eukaryotes, alternative splicing is a key step in the regulation of gene expression, with important biomedical associations. 

The recent advent of sequencing-based transcriptomics (RNA-seq) has provided a great opportunity to quantify gene expression and study alternative splicing. Several mature computational methods exist for transcript abundance and alternative isoform quantification, however all methods are geared towards static RNA-Seq experiments and are still problematic for low covered genes/ isoforms. Recently, time series RNA-seq data sets, often in combination with biotin labelling, have been generated to illustrate the kinetics of RNA synthesis and splicing, creating a need and an opportunity for isoform quantification methods that may be able to take advantage of temporal information in RNA-Seq processing. In this work, we propose a Bayesian method to jointly infer isoform proportions from RNA-seq time-series data. The method is based on a mixture model whose mixing proportions (isoform ratios) are coupled through a latent Gaussian process. This dynamical model shows efficient performance and results in lower uncertainty in inferring isoform proportions and better accuracy at low coverages. Experiments on simulated and real data sets from yeast and mouse demonstrate the potential of the method in real biological investigations. Python code is freely available at http://diceseq.sf.net

Guido Sanguinetti

"Pseudomarginal inference for Markov Jump Processes via random truncations"

Markov Jump Processes are discrete state, continuous time stochastic processes which are increasingly popular models for stochastic reaction processes. Despite their simplicity, inference and parameter estimation in these processes remains difficult, particularly for systems with infinite state space. In this talk I will present a new pseudo-marginal MCMC method which enables asymptotically exact inference also for infinite state-spaces by introducing random trunctations of the state space. Remarkably, this method is both asymptotically exact and highly computationally efficient, giving good estimates on benchmark systems in times comparable with variational approximations. 

Joint work with Anastasis Georgoulas and Jane Hillston