Dresden 2017 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
BP: Fachverband Biologische Physik
BP 50: Networks: From Topology to Dynamics I (Joint Session SOE/DY/BP)
BP 50.3: Talk
Thursday, March 23, 2017, 10:15–10:30, GÖR 226
Improving Causal Gaussian Bayesian Network Inference using Parallel Tempering — •Pascal Fieth1, Gilles Monneret2,3, Andrea Rau3, Florence Jaffrézic3, Alexander K. Hartmann1, and Gregory Nuel2 — 1IfP, University of Oldenburg, Germany — 2LPMA, CNRS 7599, UPMC, Paris, France — 3GABI, INRA, Jouy-en-Josas, Paris, France
Gene regulatory networks describe causal relationships in biological processes like signal transduction or disease mechanisms. A considerable interest exists in supporting experimental network inference by developing computational methods to infer gene regulatory networks from available gene expression data.
To infer causality within those networks from mixed, observational and intervention, data, we make use of causal orderings and Gaussian Bayesian networks. An introduction to the necessary foundations is given. In the presented framework, for a given causal ordering, the likelihood of the model network can be maximized analytically. The space of causal orderings, growing as n! for n genes, can be reliably explored via a simple Markov Chain Monte Carlo algorithm[1] for 10-20 genes only.
We show that parallel tempering helps in finding the orderings with highest maximum likelihood estimators as well as in exploring the set of alternative orderings with comparable maximum likelihood estimators for networks with >50 genes.
[1] A. Rau, F. Jaffrézic, G. Nuel, BMC Sys. Biol., 7(1):111 (2013)