Machine Learning for Personalized Medicine

Probabilistic modelling of omic time course data

by Magnus Rattray

We are building models of the kinetic processes regulating the production of mRNA using time course data from high-throughput genomic technologies. We use non-parametric Gaussian process models to represent time-varying concentrations or activities and we integrate these non-parametric models with simple mechanistic difference [2] or differential equation models [1]. Parameter estimation and model-based prediction is carried out using Bayesian inference techniques. We use these models to help understand what are the rate-limiting steps in gene expression and how these processes are regulated. 

References:

[1] Honkela, A, et al. "Genome-wide modelling of transcription kinetics reveals patterns of RNA processing delays." Proceedings of the National Academy of Sciences USA (in press, 2015: preprint at http://arxiv.org/abs/1503.01081)

[2] wa Maina, Ciira, et al. "Inference of RNA Polymerase II Transcription Dynamics from Chromatin Immunoprecipitation Time Course Data." PLoS Computational Biology 10.5 (2014).

>back to "Talks and Speakers"