Interpretation of sensitivity function values

[shoepfl]

I have a question about the actual implementation of the forward sensitivity analysis. I assume amici calculates the forward sensitivity like:

with the kth parameter of the parameter vector theta and the state vectors xi. This would however not allow to compare the values of different species with different units or different units of parameters right? So is there any normalization included?

Thanks for the feeback,

Sebastian

[FFroehlich]

I have a question about the actual implementation of the forward sensitivity analysis. I assume amici calculates the forward sensitivity like:

with the kth parameter of the parameter vector theta and the state vectors xi. This would however not allow to compare the values of different species with different units or different units of parameters right? So is there any normalization included?

Thanks for the feeback,

Sebastian

Hi Sebastian. Yes you are correct, AMICI always computes unnormalized sensitivities. For most biochemical systems, I would recommend calculating sensitivities with respect to a logarithmic scale as this is a more natural scale (for mass action kinetics log-parameters correspond to Gibbs free energies) and should also address any concerns about normalization or units.

Hope this helps,
Fabian

[shoepfl]

Thanks for the fast reply Fabian,

yes this helps. So log transformation of the parameters with
amici_model.setParameterScale(amici.ParameterScaling.log10)
should do the job to get normalized sensitivities, right?

Best,

Sebastian

[FFroehlich]

Yes, thats correct. setParameterScale expects a vector of value though, one for each parameter in the same ordering as getParameterIds (It looks like the documentation for that function is not being exported correctly)/