|
|
|
|
Signaling pathways from RNAi Data
A collaboration with the group of Michael Boutros at DKFZ Heidelberg. In RNA interference (RNAi) small double-stranded RNA molecules are designed to specifically target cellular mRNAs for degradation. The method allows to knock-down expression of candidate genes. (Visit the Nature web focus on RNA interference to view an animated tour through the process of RNAi.) Effects of RNAi interventions are measured by microarrays. We work on methods to combine information from effects of gene silencing into pathway models. Our work focusses on two topics specific to RNAi data: 
Probabilistic soft interventions:
We model data from silencing experiments in a Bayesian framework and discuss differences knock-outs.
In Markowetz et al. (2005a) we introduce a general concept of probabilistic interventions in Bayesian networks.
This generalizes deterministic interventions, which fix nodes to certain states.
Signaling pathways: The figure on the left from (Boutros et al., Dev Cell, 2002) shows effects of silencing genes in a pathway of Drosophila immune response. Columns correspond to controls and RNAi experiments, while rows show effects of such interventions on genes measured on a micorarray. The objective is to reconstruct a pathway topology from the results of interventions experiments. The effect of silencing one candidate gene on other candidate genes may not be observable on microarrays when signaling happens on protein level. In Markowetz et al. (2005b) we introduce a method to infer non-transcriptional pathway features from secondary effects of gene silencing.
Publications
|
|||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||
We propose "pushing" variables in the direction of target states without fixing them.
We formalize this idea in a Bayesian framework based on Conditional Gaussian networks.
The figure on the right shows pushing applied to a Dirichlet prior - the stronger the intervention, the more the distribution concentrates at a target value.