Why the simulated plot is different when select style between "Time" and "Percentile"?

Hi,

I do the simulate with my model, and compared the simulated value with observed data, and I want to adjuste the initial values of the parameters according to the comparison results.

But when I compared the simulated value with observed data, I find in the Property Editor, when I select "Time" STYLE, the simulated value was very close to the observed data (figure in the left); when I selece "Percentile" STYLE, imulated value was very different with the observed data (figure in the reight). So I don't know why this is happening?

This disturbs me, because when I used the model with the intial values used in the simulation step to fit my observed data, there is an error say "Optimization cannot proceed because the initial value of the negative log-likelihood is NaN. ..." In "Time" STYLE, the initial parameter looks like well, in "Percentile" STYLE, the initial parameter looks bad. How can I adjust the parameters?

Answer 1

[Priya Moorthy]

Hi,

Percentile plots are usually meant for visualize a large number of simulation results. Each individual timecourse is interpolated onto a common time vector and then the percentiles are calculated across each timecourse for each time point on the common time vector. From your plot, it looks like the common time vector used in the percentile plot doesn't include a time point near the peak, so you are seeing a smoothing effect due to interpolation. You can enter a custom common time vector in the percentile plot settings to interpolate the simulation results at the desired timepoints. In general, percentile plots are not useful for viewing single simulation results, and the time plot will give you the most precise view of the simulation data.

Best,

Priya

[Florian Augustin]

Hi voZnW,

I wanted to follow up on Priya's answer and comment on the error you are seeing in the Fit program. I have seen this error message when a proportional error model is used for response values and data that are exactly zero. Looking at your plots, your data and response seem to be 0 at time 0. We recommend removing this one data point at time 0 from your data set when proportional error models are used. If you are not using a proportional error model, then I would need more information about your fit setup to help, would you be able to share your project file?

Best,

Florian

[voZnW]

Thank you very much for your answer.

Yes, I am happy to share my project file, please see the attachments.