Making models match measurements: Model optimization for morphogen patterning networks

Abstract

Mathematical modeling of developmental signaling networks has played an increasingly important role in the identification of regulatory mechanisms by providing a sandbox for hypothesis testing and experiment design. Whether these models consist of an equation with a few parameters or dozens of equations with hundreds of parameters, a prerequisite to model-based discovery is to bring simulated behavior into agreement with observed data via parameter estimation. These parameters provide insight into the system (e.g., enzymatic rate constants describe enzyme properties). Depending on the nature of the model fit desired - from qualitative (relative spatial positions of phosphorylation) to quantitative (exact agreement of spatial position and concentration of gene products) - different measures of data-model mismatch are used to estimate different parameter values, which contain different levels of usable information and/or uncertainty. To facilitate the adoption of modeling as a tool for discovery alongside other tools such as genetics, immunostaining, and biochemistry, careful consideration needs to be given to how well a model fits the available data, what the optimized parameter values mean in a biological context, and how the uncertainty in model parameters and predictions plays into experiment design. The core discussion herein pertains to the quantification of model-to-data agreement, which constitutes the first measure of a model's performance and future utility to the problem at hand. Integration of this experimental data and the appropriate choice of objective measures of data-model agreement will continue to drive modeling forward as a tool that contributes to experimental discovery. The Drosophila melanogaster gap gene system, in which model parameters are optimized against in situ immunofluorescence intensities, demonstrates the importance of error quantification, which is applicable to a wide array of developmental modeling studies.

Systems biology derived source-sink mechanism of BMP gradient formation

“…Error between the model results and the fluorescent data are calculated via a two-step process. First, the amplitude of the P-Smad5 fluorescent-intensity data and model peak levels for free BMP are normalized as commonly done when calculating a residual with fluorescent intensity data ( Hengenius et al, 2014 ; Pargett and Umulis, 2013 ). This approximation is valid considering that (1) BMP ligands are not saturating receptors and (2) Smad5 activity is not saturated ( Figure 6G ).…”

Deciphering the Signaling Network Landscape of Breast Cancer Improves Drug Sensitivity Prediction

“…We tested the fits of models to the data using the root mean square error (RMSE); the lower the RMSE value the better the fit (Hengenius et al, 2014). The models fit the data very well with an average error of only 5%, which is in the same range as biological replicate average error of 4% ( Fig.…”

Image-based modelling of organogenesis

“…Which metric is most suitable depends on the type of experimental data and the model [14,71,72]. Thus, only very little data is truly quantitative, such as the strength of the growth fields discussed above.…”