A linearized constraint-based approach for modeling signaling networks.

With the unparalleled increase in the availability of biological data over the last couple of decades, accurate and computable models are becoming increasingly important for unraveling complex biological phenomena. Past efforts to model signaling networks have utilized various computational methods, including Boolean and constraint-based modeling (CBM) approaches. These approaches are based on solving mixed integer linear programs; hence, they may not scale up for the analysis of large networks and are not amenable for applications based on sampling the full spectrum of the solution space. Here we propose a new CBM approach that is fully linear and does not involve integer variables, thereby overcoming the aforementioned limitations. We describe a novel optimization procedure for model construction and demonstrate the utility of our approach on a reconstructed model of the human epidermal growth factor receptor (EGFR) pathway, spanning 322 species and 211 connections. We compare our model's predictions to experimental phosphorylation data and to the predictions inferred via an additional Boolean-based EGFR signaling model. Our results show high prediction accuracy (75%) and high similarity to the Boolean model. Considering the marked computational advantages in terms of scalability and sampling utilization obtained by having a linear mode, these results demonstrate the potential promise of this framework for the study of cellular signaling.