Dynamic basis pursuit regularization for complex biochemical pathway identification

The availability of both reliable parameter (kinetic constant) estimates and knowledge about sensitive pathway interactions are still limiting steps in the analysis of biochemical signal transduction pathways. This paper investigates feature selection/model reduction in biochemical pathways by examining parameter sensitivity using basis pursuit regularization. A 1-norm model complexity measure allows model structures to be ranked in a continuous manner. In particular, this paper analyzes the limitations associated with collocation-based approaches to pathway parameter locus identification which transform dynamic parameter estimation into a simple algebraic problem. The bias associated with these approaches can be overcome using a dynamic basis pursuit regularization approach which is developed, analyzed and compared with collocation approaches.