Reconstruction of flexible gene-protein interaction networks using piecewise linear modeling and robust regression

Abstract

In this study we will focus on piece-wise linear state space models for gene-protein interaction networks. We will follow the dynamical systems approach with special interest for partitioned state spaces. From the observation that the dynamics in natural systems tends to punctuated equilibria, we will focus on piecewise linear models and sparse and hierarchic interactions, as for instance described by Glass, Kauffman, and de Jong. Next, the paper is concerned with the identi_cation (a.k.a. reverse engineering and reconstruction) of dynamic genetic networks from microarray data. We will describe exact and robust methods for computing the interaction matrix in the special case of piecewise linear models with sparse and hierarchic interactions from partial observations. Finally, we will analyze and evaluate this approach with regard to its performance and robustness towards intrinsic and extrinsic noise.