Advanced_Examples_For_Parameter_Optimization

140 views
Skip to first unread message

Niels Zondervan

unread,
Aug 26, 2015, 10:21:12 AM8/26/15
to COPASI User Forum
Dear all,

I am working on a metabolic model for which both steady state and dynamic times series data (C12/C13 ratio's and MS absolute and relative data).
With this abundance of data also come some challenges.
I have a hard time finding information about the following:
  1. Comparisons or advices on which parameter estimation methods works better for for steady state data, which work well for times series, which work well for C12/C13 ratios.
  2. The optimal order for parameter optimization, first fit to steady state data and afterwards optimize with time series or the other way around (I do think the order matters in avoiding for example local minima).
  3. Scaling, when using relative data instead of absolute concentrations how to let COPASI take this into account
I am fairly certain such information is available but it has eluded all my search attempts so far. Any suggestions or references on these three problems are very welcome.
Basically what I really am searching for is a more advanced tutorial on parameter optimization and COPASI. I found a few tutorials on COPASI which were illustrative but to basic to answer these questions.

Best,

Niels


Sluka, James Patrick

unread,
Aug 26, 2015, 12:18:02 PM8/26/15
to copasi-u...@googlegroups.com

Niels

 

I can’t address everything you asked but here are a few thoughts.

 

If the fitting is done “properly” then the order shouldn’t matter. Indeed, I would do it both ways until the two ways give the same result.

 

In my experience the gradient based fitting methods don’t work well. If the model is pretty simple (and has no cycles) and if the original estimates of the parameters are pretty good then those methods work OK and are very fast. In general though you can’t escape local minima using those methods unless you use different initial parameters, a tedious process. A more reliable approach is to use tools like Simulated Annealing and Particle Swarm. Those methods are much slower but are much less sensitive to local minima. The difference in run time are significant and the Simulated and Annealing and Particle Swarm calculations might take an hour or more (vs. a minute or so for the other methods) but in practical terms these search methods will get you to a better answer quicker.

 

Jim

--
You received this message because you are subscribed to the Google Groups "COPASI User Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to copasi-user-fo...@googlegroups.com.
To post to this group, send email to copasi-u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/copasi-user-forum/afd1883b-e9d8-4b27-9f64-abfd1feab7b5%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Stefan Hoops

unread,
Aug 26, 2015, 1:55:22 PM8/26/15
to copasi-u...@googlegroups.com
Hello Niels,

> 1. Comparisons or advices on which parameter estimation methods
> works better for for steady state data, which work well for times
> series, which work well for C12/C13 ratios.

According to the no free lunch theorem there is no answer to which is
the best method for a particular problem. I personally would start with
Particle Swarm or Scatter Search. If you have a good result and want to
refine it use a local method like Levenberg Marquardt or Hooke and
Jeeves.

> 2. The optimal order for parameter optimization, first fit to
> steady state data and afterwards optimize with time series or the
> other way around (I do think the order matters in avoiding for
> example local minima).

If you do not have a special reason to fit the parameters first for one
experiment and than for another I suggest that you fit all at the same
time by adding multiple experiments. The order in which they added does
not have any effect on the calculation.

> 3. Scaling, when using relative data instead
> of absolute concentrations how to let COPASI take this into account
>

You can use assignments to calculate the relative amounta in COPASI
and fit those instead of the concentrations.

Thanks,
Stefan




--
Stefan Hoops, Ph.D.
Research Assistant Professor
Virginia Bioinformatics Institute
Virginia Tech
1015 Life Science Circle (0477)
Blacksburg, Va 24061, USA

Phone: (540) 231-1799
Fax: (540) 231-2606
Email: sho...@vbi.vt.edu

Niels Zondervan

unread,
Aug 30, 2015, 7:26:27 AM8/30/15
to COPASI User Forum
Hello Jim,

Thank you so much for the advise. I do have a model without cycles, currently only 22 reactions but I plan to expand the network with more pathways.
Computing power and computing time is not much of an issue, the aim is to find the best possible set of parameter with the data I have.
I plan to first balance my parameters using a Bayesian parameter balancing approach to find consistent combinations of kinetic parameters which get balanced with steady state concentrations. I will follow your and Stefan's advice and use the Particl Swarm algorithm, I will also have a look at  Scatter Search and Simulating Aneeling algorithm. It surprises me that the order in which parameter estimation is performed does not matter, it shows I still have to improve my understanding of these algorithms.

Thanks,

Niels

Niels Zondervan

unread,
Aug 30, 2015, 7:48:03 AM8/30/15
to COPASI User Forum
Hi Stefan,

I will follow your and  Jim's advice and use the Particle Swarm algorithm, I will also have a look at the Scatter Search.
I just searched for information on the use of assignments of a species quantity in Copasi, but have some problems understanding it because I cannot find a good example.
Is it describing a Species quantity with an ODE times some scaling factor?
If you know of an example, tutorial  or paper where they use assignments this would really help a lot.

Thanks,

Niels

Stefan Hoops

unread,
Aug 30, 2015, 8:29:15 AM8/30/15
to copasi-u...@googlegroups.com
Hello Niels,

If you measured the ratio of S1/S2 in your experiments and have species
S1 and S2 calculated in your model you need to create a global quantity
R of type ASSIGNMENT and calculate it as the ratio of the
concentrations of S1 and S2. To select the ration use the COPASI icon
next to the expression edit widget.

Thanks,
Stefan


On Sun, 30 Aug 2015 04:48:03 -0700 (PDT)
Niels Zondervan <nazon...@gmail.com> wrote:

> If you know of an example, tutorial  or paper where they use
> assignments this would really help a lot.



Niels Zondervan

unread,
Aug 30, 2015, 10:29:20 AM8/30/15
to copasi-u...@googlegroups.com
Hi Stefan,

Thanks for the example, I understand it now.

Thanks,
Niels



--
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/IIOCaBpdahw/unsubscribe.
To unsubscribe from this group and all its topics, send an email to copasi-user-fo...@googlegroups.com.

To post to this group, send email to copasi-u...@googlegroups.com.

Niels Zondervan

unread,
Sep 11, 2015, 10:00:21 AM9/11/15
to COPASI User Forum
Hi Stefan,

I implement the ratios today.I created global variables called R1, R2  etc. of Simulation Type 'assignment',  I selected the substrates via the select object button (Copasi symbol -> species, transient Concentrations and then added my species. The Expression looks like {[S1]}/{[S2]} in code view.
I still have a problem seeing how this can be used with my data since I have C13/C12 ratio's per compounds as measurements (S1-c13/S1-c12,
S2-c13/S2-c12 etc.). I think I understand what you mean now, but unfortunately it is not applicable in my situation since I do not have S1/S2 ratio's. If I misunderstood and your method can also be used for my data, please let me know.
My current plan is to calculate the fluxes based on the C13/C12 ratio's per compound using the tool 13cFlux2 and then fit the fluxes using COPASI.

Thanks,

Niels

Pedro Mendes

unread,
Sep 11, 2015, 10:18:02 AM9/11/15
to copasi-u...@googlegroups.com
Hi Niels,

I think your proposed solution may be the easiest (fitting fluxes after
you have calculated them with another tool).

best wishes
Pedro
> Email: sho...@vbi.vt.edu <javascript:>
>
> --
> You received this message because you are subscribed to the Google
> Groups "COPASI User Forum" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to copasi-user-fo...@googlegroups.com
> <mailto:copasi-user-fo...@googlegroups.com>.
> To post to this group, send email to copasi-u...@googlegroups.com
> <mailto:copasi-u...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/copasi-user-forum/f264b20f-efff-4f59-b23b-5538bb8eb1a6%40googlegroups.com
> <https://groups.google.com/d/msgid/copasi-user-forum/f264b20f-efff-4f59-b23b-5538bb8eb1a6%40googlegroups.com?utm_medium=email&utm_source=footer>.
> For more options, visit https://groups.google.com/d/optout.

--
Pedro Mendes
Professor of Computational Systems Biology
http://www.comp-sys-bio.org
School of Computer Science
Manchester Centre for Integrative Systems Biology
University of Manchester

Manchester Institute of Biotechnology
131 Princess Street
Manchester, M1 7DN, U.K.

Stefan Hoops

unread,
Sep 11, 2015, 10:23:49 AM9/11/15
to copasi-u...@googlegroups.com
Hello Niels,

OK, if I understand you correctly you measured the ratios of
S1-c13/S1-c12, S2-c13/S2-c12 etc

In you model you have only S1, S2 etc. which are the totals i.e.,
S1-c13 + S1-c12.

Are you assuming that ..-c13 is fixed during the process or may it
change? If it fixed you just create the appropriate species and calculate as
the ratios. If not you need to model how it changes differently
than ..-c12.

Thanks,
Stefan



On Fri, 11 Sep 2015 07:00:21 -0700 (PDT)
> > Email: sho...@vbi.vt.edu <javascript:>

Niels Zondervan

unread,
Sep 14, 2015, 7:37:38 AM9/14/15
to COPASI User Forum
Hi Stefan,

I have indeed C13 divided by C12, so a ratio per compound. The measurements are from a C13 glucose and show how quickly the glycolysis is saturated by C13 (shown by the ratio C13/C12 for each compound over time). I am searching for the most straightforward way to use this data e.g. by calculating fluxes and use these fluxes for parameter optimization. Most likely I first need to calculate fluxes using other software like OpenFlux of 13CFlux2. I hoped Copasi already had some build in methods to calculate fluxes using C12/C13 measurements, but I guess that is not the case.

Niels
Reply all
Reply to author
Forward
0 new messages