Your opinion about FitMultiCell.
27 views
Skip to first unread message
Эрвин Визард
Dec 22, 2021, 5:47:10 AM12/22/21
to Morpheus users
Dear Colleagues,
Sometimes a parameter values in Cellular Potts Modeling have effective sense or can not be found directly in literature.
For this reason, it becomes especially important be ready to use some algorithms and tools for parameter estimation.
At least, this makes it possible to give a ready answer at conferences or when writing articles :)
In our pratice in Pharmacometricts we use parameter estimation for ODE systems very often.
But using, for example, such algorithms like FOCE or stochastic SAEM for non-ODE models can be very problematic. Also this algorithm are associated with corresponding tools.
In the case of Morpheus, as I understand, we have such possibility to use more sophisticated approach - Approximate Bayesian Computation.
I have several questions about FitMultiCell for Morpheus:
1) Is this actual information about the coupling of Morpheus and FitMultiCell?
2) Can we run parameter fitting procedure from Morpheus or should operate only within FitMultiCell?
Sometimes a parameter values in Cellular Potts Modeling have effective sense or can not be found directly in literature.
For this reason, it becomes especially important be ready to use some algorithms and tools for parameter estimation.
At least, this makes it possible to give a ready answer at conferences or when writing articles :)
In our pratice in Pharmacometricts we use parameter estimation for ODE systems very often.
But using, for example, such algorithms like FOCE or stochastic SAEM for non-ODE models can be very problematic. Also this algorithm are associated with corresponding tools.
In the case of Morpheus, as I understand, we have such possibility to use more sophisticated approach - Approximate Bayesian Computation.
I have several questions about FitMultiCell for Morpheus:
1) Is this actual information about the coupling of Morpheus and FitMultiCell?
2) Can we run parameter fitting procedure from Morpheus or should operate only within FitMultiCell?
3) How should user of FitMultiCell/Morpheus mark variable of interest to compare it then with some experimental knowledge and to calculate some type of discrepansy?
---
Thank you for your help!
Thank you for your help!
Message has been deleted
Lutz Brusch
Dec 26, 2021, 5:13:19 AM12/26/21
to Morpheus users
Dear Эрвин,
true, since CPM is inherently stochastic and observables, trajectories etc. form ensembles at fixed parameters, special care must be taken for parameter estimation - this is what FitMultiCell is doing for you.
1) Please see the documentation on how to couple Morpheus and FitMultiCell: https://fitmulticell.readthedocs.io
You may first want to reproduce the examples from https://fitmulticell.readthedocs.io/en/latest/example.html and then swap an example for your actual problem.
You may first want to reproduce the examples from https://fitmulticell.readthedocs.io/en/latest/example.html and then swap an example for your actual problem.
2) FitMultiCell wraps around Morpheus. You have to first develop your model in Morpheus (GUI) and make sure it works qualitatively. This model.xml file will then be used in the Python code of FitMultiCell by automatically manipulating parameter values and starting the Morpheus simulator (not GUI) many times as needed. So the parameter estimation run itself is NOT started from Morpheus (GUI) but by running the Python code of FitMultiCell. Note, plain parameter sweeps (or scans) are available directly inside Morpheus, but these merely flood you as the user with all the simulation data and do not pick better parameters (which FitMultiCell does).
3) You need to tell FitMultiCell the 2 interfaces to your model.xml (in addition to the experimental data set). This declaration of the interfaces is supported by the user-friendly PEtab format (see link under 1).
(i) The parameters that shall be optimized need to be given by their XPath string. This XPath can be looked up in the Morpheus GUI by right-click on the parameter name, which opens the context menu, then click the third line="Copy XPath". Then insert these strings into the PEtab problem description (or the Python code directly).
(ii) The observables that shall be compared to experimental data need to be logged by Morpheus/Analysis/Logger or can be computed post-hoc in Python from the output data in the logger file(s). The Python code of FitMultiCell imports selected columns from the logger file(s) as selected by the observables in the PEtab specification. Another PEtab specification then lets you pair the logged/computed simulation observables to the loaded experimental data and choose various summary statistics and (weighted cumulative) distance metrics (across multiple distances).
(i) The parameters that shall be optimized need to be given by their XPath string. This XPath can be looked up in the Morpheus GUI by right-click on the parameter name, which opens the context menu, then click the third line="Copy XPath". Then insert these strings into the PEtab problem description (or the Python code directly).
(ii) The observables that shall be compared to experimental data need to be logged by Morpheus/Analysis/Logger or can be computed post-hoc in Python from the output data in the logger file(s). The Python code of FitMultiCell imports selected columns from the logger file(s) as selected by the observables in the PEtab specification. Another PEtab specification then lets you pair the logged/computed simulation observables to the loaded experimental data and choose various summary statistics and (weighted cumulative) distance metrics (across multiple distances).
Please let us know at https://gitlab.com/fitmulticell/fit/-/issues how it worked for your problem and which steps need to be better explained in the docu or need to be covered by additional examples.
Best wishes,
Lutz
Эрвин Визард
Dec 28, 2021, 5:04:00 AM12/28/21
to Morpheus users
Dear Lutz,
Thank you for your detailed answer!
I hope in the near future I will be able to adapt my task for use in conjunction with FitMultiCell.
Thank you for your detailed answer!
I hope in the near future I will be able to adapt my task for use in conjunction with FitMultiCell.
My current problem is that I'm not a master in Python language. We use R as deafault language for data preparation and processing.
Therefore, learning and operating with this package can take an indefinite time...
Thank you and Happy Upcoming New Year!
воскресенье, 26 декабря 2021 г. в 13:13:19 UTC+3, Lutz Brusch:
Reply all
Reply to author
Forward
0 new messages