Transcript Hackathon2013_Roth_OptimisingReducedModels

Optimization and testing of a reduced
model of pyramidal neurons
Arnd Roth
Wolfson Institute for Biomedical Research
University College London
How do neurons transform synaptic
inputs into action potential output?
What are the functional compartments
in neurons?
Backpropagation of action potentials
Stuart & Sakmann, 1994
Dendritic spikes in vivo
Helmchen, Svoboda,
Denk & Tank, 1999
Local dendritic spikes in vivo
Smith, Smith, Branco & Häusser, 2013
BAC firing in layer 5 pyramidal cells
Larkum, Zhu &
Sakmann, 1999
Critical frequency for calcium electrogenesis
Larkum, Kaiser &
Sakmann, 1999
Constraining passive parameters
in compartmental models
Roth & Häusser (2001)
„Raw“ and „core“ model parameters
Turning bad space clamp into a virtue
Schaefer, Helmstaedter, Sakmann & Korngreen, 2003
Dendrotomy and „pinching“
Bekkers & Häusser, 2007
Dendrotomy and „pinching“
change somatic AP waveforms
Detailed and reduced models
Traub et al., 1991
Pinsky & Rinzel, 1994
Case study: a reduced compartmental
model of a layer 5 pyramidal neuron
Bahl, Stemmler, Herz & Roth, 2012
Verification of passive properties
Pinching in a layer 5 pyramidal neuron
Model generalization
f/I curves before and after pinching
BAC firing
Energy efficiency of action potentials in the Hodgkin-Huxley model
Hodgkin´s notion (1975):
„In the squid axon ... the entry of sodium
and loss of potassium per impulse are
3 to 4 times the theoretical minimum of
Cm UAP.“ (Charging a pure capacitor)
Percentage of energy expended on
neuronal signalling in rat neocortex
Redistribution
of ions:
Na+-K+-ATPase
Hodgkin & Huxley J. Physiol. (London)
(1952d); Figs. 17 and 18
Attwell & Gibb Nature Reviews Neuroscience
(2005); Fig. 2 mod.
Transmembrane ion fluxes during a mossy fiber bouton
action potential at 37 °C
Charge separation
Transmembrane ion fluxes during a mossy fiber bouton
action potential at 37 °C
Alle, Roth & Geiger, 2009
Challenges
•
Nonlinearity and large number of free parameters
present a difficult optimization problem
•
Which data and which distance functions can
constrain which type of model?
•
Tradeoff between flexibility and identifiability
•
Access to high-quality, internally consistent,
annotated data sets
What we need
•
Test suites to objectively compare different (classes
of) models (-> Open Source Brain)
•
Standardized, portable model description (-> NeuroML)
•
Simulator-independent model construction (e.g.
neuroConstruct) and optimization (e.g. MOO)
•
Integration of data from electrophysiological and
optical recordings with morphological and molecular
biology data
•
Objective methods for dimensionality reduction