Inconsistent Sensitivity Analysis

[Tyler Nicholas Simmons]

Hello everyone,

I have created a deterministic ODE model in COPASI. I have recently learned that every time I open COPASI and run the model, the sensitivity analysis provides me different results. Other tasks yield the same results no matter when I run it. Only the sensitivity analysis does not. A majority of the unscaled and scaled sensitivity analysis is consistent, but for certain species, the coefficients are always different. When looking at the summarized sensitivities, there is no consistency whatsoever, different every time. 

If anyone has encountered this problem or knows what's happening, please let me know.

Best,

[Frank Bergmann]

Hello Tyler, 

this will be difficult to answer without seeing the model. So the only thing i could guess at so far, would be numerical instabilities, resulting in slightly different unscaled values. And of course, once they are scaled with the parameter values, the difference might compound for the scaled sensitivities. 

best

Frank

[Tyler Simmons]

Thank you for this answer. I am still confused about the methodology behind this analysis. How would numerical instability result in different values every time. I get that they may be different and may compound when scaled, but how is the initial computations with the same equation and same parameters yielding different numbers each time?

--
You received this message because you are subscribed to a topic in the Google Groups "COPASI User Forum" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/copasi-user-forum/lkDn3P03Xog/unsubscribe.
To unsubscribe from this group and all its topics, send an email to copasi-user-fo...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/copasi-user-forum/028eaa16-5633-4d54-a7d6-d0879020364en%40googlegroups.com.

--

Tyler Simmons

Biophysics Graduate Student

University of Maryland

College Park, MD

[Frank Bergmann]

COPASI uses finite differences, to compute the sensitivities. For each cause a permutation is applied and the selected subtask run twice to compute the difference. To answer where the differences come from, we'd have to know more like what subtask was selected. How different are the results you observe? Are the differences on the sensitive parameters? What role do these parameter play in the model and the like. 

That's why I said, that so far we can only guess. 

best

Frank