BlueBrain
Pre made not uniform model
Hi,
I found this new study very nicely made a AIS model. https://senselab.med.yale.edu/ModelDB/showmodel.cshtml?model=267355#tabs-2
I am thinking about if can I put this onto the L5PC example, replacing the AIS.
In my experience, using the soma recording to fit AIS parameters is very difficult. If this AIS model is good, It can make the model more realistic.
What would be the best way to do so? I am assuming we have to modify the evaluator?
Thank you so much!
Yuan-Ting Wu
Hello @bhmagic,
I am not sure I understood your question, so let me know if that doesn't help:
If you wish to use their exact parameters (https://senselab.med.yale.edu/ModelDB/showmodel.cshtml?model=267355&file=/aisL5/Cell%20parameters.hoc#tabs-2) you can modify the parameters.json of the L5PC example to match theirs and copy their mechanisms in the L5PC mechanisms directory. If your question was about the distribution of the parameters, then indeed it might be that what they did is not possible in the formalism of BluePyOpt. In that case, the best would be to code a new parameter scaler that fit your need (https://github.com/BlueBrain/BluePyOpt/blob/master/bluepyopt/ephys/parameterscalers/parameterscalers.py).
For the fitting, if you do not have AIS specific data, you will have to rely on somatic data to fit the AIS or axonal parameters. However, if you do have data, from experiments or from the literature, you do not have to rely only on somatic recordings. You can define recordings for non-somatic location in BluePyOpt, see for example the classes https://github.com/BlueBrain/BluePyOpt/blob/master/bluepyopt/ephys/locations.py
My question is indeed about the distribution of the parameters (within a component). I am thinking maybe forcing it to be uniform. I will explore further!
Note that you can define custom distributions as it is the case in the L5PC example for the Ih channels: https://github.com/BlueBrain/BluePyOpt/blob/master/examples/l5pc/config/parameters.json#L144, but they can only be distance-dependent.