Benchmark-Models-PEtab
Benchmark-Models-PEtab copied to clipboard
Benchmarking-Initiative
Add Fribourg et al. (2014)
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4012420/
SBML: https://www.ebi.ac.uk/biomodels/BIOMD0000000529#Overview
Estimated complexity of implementation: low
@dweindl @LeonardSchmiester @yannikschaelte The model and the dataset is very interesting, however, we have observables y(t,u) = x_i(t,u)) / x_i(T,U)) with the time point T \geq t. For some measurements the normalisation is even done with respect to another simulation condition u \neq U. In my opinion we should get more or less the same of we introduce a scaling constant y(t,u) = s_i*x_i(t,u)). The fitting problem is not exactly the same, but as for the case above only information about relative changes would be considered. What do you think?
I agree that the fitting problem would be different. Since x_i(T,U) is based on the simulations which is not available for (t,u), the scalings s_i would need to estimated. Indeed, this also puts info on a relative scale, so the problem would be similar. If necessary, it should be possible to replicate the exact problem in pyPESTO modifying the AMICI outputs manually. This gets a bit nasty for derivatives though ...
Agreed. I think it would be fine to implement it as Jan suggested with scalings and accept that it is not 100% the same optimization problem.
If the goal is not replicating the problem exactly, that works for me too. Should be clearly stated somewhere. Implementing the original problem would require some rather complex(?) PEtab adaptations.
Should be clearly stated somewhere.
Agreed!
I think these slight adaptations should be fine. Should be like using a slightly different statistical model.