Abstract
Characterization of signal execution dynamics within complex biochemical networks is a highly challenging problem yet necessary to understand how cells process signals and commit to a biological phenotype. The mechanistic interpretation of experimental results can often be misinterpreted due to limited available data or the need for an unrealistic number of experimental measurements. Mathematical models of biochemical networks have emerged as an alternative to complement experiments and explore signal execution mechanisms. However, traditional computational methods require either detailed knowledge of model parameters or sufficient data to calibrate models to experiments, both of which can be difficult to obtain. To address this challenge, we have taken a probabilistic approach to the analysis of network-driven biochemical processes. In this work, we apply a Bayesian multimodel inference formalism to identify how relevant pathways and subnetworks contribute to the overall mechanism of a biochemical signaling network. We focus this approach on the signal execution pathways of mammalian extrinsic apoptosis. We study the effect of changing concentrations of key apoptosis regulators such as XIAP, which governs the phenotypic mode of apoptotic execution, either via the mitochondria independent (Type I) or dependent (Type II) pathways. Several hypotheses were generated regarding (i) differential pathway regulation by XIAP; (ii) apoptotic signal arrest through the caspase-only pathway; and (iii) the primary signal amplification concomitant with XIAP inhibition leads to signal amplification via mitochondrial involvement. Our findings substantiate the use of probabilistic and multimodel inference-based approaches for the hypotheses exploration regarding the mechanisms of signal execution dynamics. We expect that these approaches could help identify key pathways in complex networks and, in turn, accelerate testable hypothesis generation.
Author summary Signaling dynamics within complex biochemical networks are remarkably difficult to characterize. Mathematical models are often used, in conjunction with experimentation, to explore the effect of changes in regulatory proteins on signal propagation. However, Mathematical models are often limited by a lack of knowledge regarding reaction rates and the lack of available data to calibrate them. To overcome this issue, we have taken a probabilistic approach over parameter space that enables reaction topology and protein concentration exploration within a Bayesian evidence context. The mechanisms of signaling dynamics and the effects of perturbations in protein concentrations can thus be hypothesized in the absence of explicit reaction rates. The method is demonstrated on a model of the extrinsic apoptosis network to explore the key drivers of Type I vs Type II signal execution. Key regulators that govern the signaling phenotype of this network are modulated and changes in the apoptotic signal through the network are analyzed. In addition to supporting established experimental results, we generate novel hypotheses regarding phenotypic control and the roles of various components of this system are made.