I downloaded some of the all-active biophysical models of interneurons from the Allen Cell Types Database. The models specify the biophysics of apical dendritic sections, but none of the reconstructed morphologies have apical dendrites. I wonder why the biophysics of the apical dendrites is defined and whether perisomatic models are preferable to all-active models with a higher explained variance ratio.
Would you mind sharing the specimen IDs of the all-active interneuron models you are looking at? I can take a look to see why there are apical dendrite parameters for those cells.
You could certainly consider the perisomatic models, as well - the explained variance ratio is a good metric to start with for performance on noise stimuli, and you could also look at the model behavior for other stimuli (e.g., hyperpolarizing steps, different duration depolarizing steps, etc) to see which one seems better for your application.
These are the specimen IDs: 485466109, 474626527, 466664172, 475202388, 321707905, 475549334, 464198958, 475888653, 485058595, 484635029, 476686112, 333785935, 487667205,
341442651, 476104386, 475057898, 482690728, 313862167.
This is a good catch! Those definitely shouldn’t be there, but also shouldn’t actually cause any issues when instantiating models - the apical parameters will simply be ignored if there are no apical dendrites. My best guess is that the model fitting by default included all possible parameters in the ‘genome’, in which case apical parameters that have no effect on model behavior would simply end up with random values (instantiated randomly and mutating randomly during the optimization).
Thank you! Yes, I haven’t found any issues so far.