parameterization

Approach in GeneNetWeaver

In GNW, parameters are sampled randomly from predefined ranges. These parameters include

1. mRNA transcription rates
2. mRNA degradation rates
3. protein translation rates
4. protein production rates
5. Hill function thresholds
6. Hill function coefficients
7. weights in logic function ($α$)

We find that sampling in this manner causes a lot of numerical problems because there are no guarantees on the numerical integration.

Approach in Dynverse

In Dynverse, the various parameters except the threshold in the Hill functions are samples once, and are set uniformly for all the variables. In order to get the desired trajectories, Dynverse stores the thresholds $k$ for the various network configurations, and these are read from file. The thresholds in the Hill functions take the form maxprotein/2/strength (see dynverse implementation), where the strength term represents the strength of the regulatory interaction the protein is a part of. All of this is hardcoded in tables.

Our approach

Currently, we prefer to carry out the synthetic data generation with as much automation as possible. We currently fix all parameters. Even after fixing $k$s to 10, we see bifurcating trajectories,