Personal tools
You are here: Home Events ANC Workshop: Emilia Wysocka, Chair: Matt Graham

ANC Workshop: Emilia Wysocka, Chair: Matt Graham

— filed under:

Dimensionality reduction of rule-based simulation results

What
  • ANC Workshop Talk
When Apr 12, 2016
from 11:00 AM to 12:00 PM
Where 4.31/4.33
Add event to calendar vCal
iCal

The subject of my talk will be a project initiated during Complex Systems Summer School in Santa Fe (NM, USA). We were interested in tackling the curse of dimensionality  with a dimension reduction technique that helps identify irrelevant variability and understand critical processes underlying high-dimensional datasets.

We used a method based on optimizing an information-theoretic objective, multivariate mutual information(Correlation Explanation, CorEx;  paper: http://arxiv.org/abs/1410.7404 https://github.com/). In principle, this method searches for a set of latent variables that best explain correlations in a dataset.

The particular motivation comes from a complex simulation results of rule-based (RB) dynamic models. Rule-based languages, such as Kappa or BioNetGen, model the combinatorial explosion of cell signalling systems. In contrast to the other modelling techniques, in RB models the system emerges with time, often showing unpredictable behaviour arising from elementary reaction rules. Even though provided with visualization tools for static and causal analysis (graphs and timeseries data), one has to resort to a self-assembled battery of heuristic tests to unfold the complexity of results. Hence, the throughput of in-silico observations, derived from simulations of large systems, may hinder or even prohibit a thorough analysis.

I will present a preliminary analysis of the method with new datasets. The datasets are results obtained with a RB model of the phosphosites regulation of DARPP-32, the dopamine and cAMP-regulated phosphoprotein of 32 kDa. DARPP-32 is a robust integrator of dopamine and glutamate signals in striatum and it’s phosphorylation sites are important in relation to drugs’ actions. The RB model is an implementation of ODE model built by Fernandez et al. 2006 (http://europepmc.org/)