Query over using sensitivity Analysis

[aryaman3900]

Hi @himoto ,

I am doing sensitivity analysis for my EGFR model and am facing an unusual thing. I have to do parametric sensitivity analysis with two conditions: EGFR concentration 1nM and 10nM, and am doing the same by three ways:

1. In conditions (in observable.py) just entered 1nM
2. In conditions (in observable.py) just entered 10nM
3. In conditions (in observable.py) entered both 1nM and 10nM

I have attached the graphs for your reference.

Surprisingly the results change in attempt 3 when I enter both the concentrations simultaneously, in attempt 2 with 10nM the change for the 1st parameter was over 100 (as shown in Fig.2) but this does not happen in Fig.3.

I would like to mention that the simulation results stay consistent in all three attempts but could you please tell me why does this happen with parametric sensitivity?

[himoto]

Hi @aryaman3900 ,

Thank you for your inquiry.

I would like to ask you this first: can you please perform sensitivity analysis with setting options['overwrite'] to True?

run_analysis(model, target='parameter', options={'overwrite': True})

I would like to know whether you will get the same results with the code above.

[aryaman3900]

Yes, this solves my query. Thank You very much.

[himoto]

Hi @aryaman3900 ,

Great to hear that!
Since the computation of sensitivity coefficients is time-consuming, current BioMASS tries to use the result: sensitivity_coefficients/ if it exists. This will let users modify visualization options again and again without computing coefficients. However, it is now difficult to detect any changes in observable.py in a model. I think this is the cause of the problem. When you modify any code in observable.py, please remove the previous calculation result via setting options={'overwrite': True}).

[aryaman3900]

Sure, I see how BioMASS works just looking from another perspective it does seems really a nice way by the program to make the computation a little less expensive. Thanks, @himoto for explaining this to me, I will ensure to use it the correct way.

Btw just a curious observation, I noticed in the documentation that both optimize and run_analysis functions have overwrite as an option but there is no such thing in run_simulation. Could you please tell what is the reason for it, does the same computational cost principle doesn't applies here?

[himoto]

@aryaman3900 ,
That's a good point.
I would like to explain why optimize and run_analysis have 'overwrite' options but run_simulation does not.
As you might know, the number of iteration for simulating a model is very large to estimate parameters or calculate sensitivity coefficients, and therefore time-consuming. To prevent unexpected deletion of estimation results or coefficients, these function have 'overwrite' options.
To my mind, run_simulation just run simulation and visualize simulation results, it does not take time and that is why I do not implement 'overwrite' option in this function.

[aryaman3900]

Sure I see. Thank You very much for clearing my query. It is very interesting to know how much has been thought at each step. Thanks!