Main page
Author's kit
Registration
Program
Organization
Venue
More information
wcsb06@cs.tut.fi
Hosted by
TUT CSB group
Abstract --- Korbinian Strimmer, Department of Statistics, Ludwig-Maximillian University, Munich, GermanyStein-Type Regularized Inference for Complex Biological Models Understanding complex biological networks on a whole-genome scale is a central objective of systems biology. However, the increasing post-genomic information flood offers substantial challenges for the systems analysis of genomic data. In my talk I focus on methodological problems related to modeling, inferring and simulation of complex networked systems. A key issue is the fit of high-dimensional models with many parameters (which correspond to genes, kinetic parameters, network edges, etc.) to genomic data that are typically are sampled from only few individuals. In order to deal with this "small n, large p" data situation we have developed an approach to Stein-type shrinkage estimation for the complex high-dimensional models encountered in systems biology. This procedure is computationally very cheap (in comparison to regularized inference based on as penalized likelihood or Bayesian procedures) and thus is ideal for the large genomic and proteomic data sets. Nevertheless, the proposed approach is statistically highly efficient. Specifically, we have applied this method to infer large scale linear graphical models, such as graphical Gaussian models, structural equations models, and vector autoregressive models from gene expression data, to describe the network-like dependencies among genes. |