Collective parameter estimation of related models with an initial stability constraint

Parameterisation of dynamic biochemical network models is a challenging aspect of systems biology. Especially when the parameter space is large and data is semi quantitive but comparable across different experimental conditions.

Here, we present a set of command line tools utilising Pycotools (COPASI) that leverages the power of high-performance computing to facilitate parameter estimation of large models with many unknown parameters. In particular, we expand upon the abilities of Pycotools to address two particular issues. Firstly, the difficulty of constraining a model’s parameterisation to assume the system begins in a steady state (prior to a perturbative stimulation). And secondly, parameterising against relative quantitative time series data that have no absolute scale.

Our software operates on the SLURM workload manager system and can be applied to any parameter estimation against time series data produced by applying a single perturbation at time zero to an equilibrated system.

We validate that our technique can produce a parameterised model of the MTOR (mechanistic target of rapamycin) network based on semi-quantitative time-series data from 2 breast cancer cell lines, stimulated with insulin and amino acids. We also show our model can make reasonable predictions on distinct signaling dynamics in one breast cancer cell line based on the other by adjusting the initial protein quantities only.

In conclusion, models should fit both the initial steady state and the dynamics following stimulation, given that stabilising systems prior to stimulation is a common experimental protocol in signaling research. By expanding standard tools, commonly used in the field, we have developed a widely applicable method, which can easily be evaluated and is amenable to wide general use.