| EFFECTS OF REALISTIC 3D NEURON MORPHOLOGY ON THE STABILITY
AND ROBUSTNESS OF A HOPFIELD-STYLE NETWORK MODEL OF WORKING MEMORY
|
| P.Coskren2,3*; J.I.Luebke4;
A.B.Rocher1,3; P.R.Hof1,3; S.L.Wearne1,2,3
|
| 1. Neuroscience, 2. Biomathematical Sci., 3. CNIC, Mt.
Sinai School of Medicine, New York, NY, USA |
| 4. Psychiatry, Boston Univ., Boston, MA, USA |
|
A recurrent neural network can simulate associative working
memory when each neuron's firing rate is considered as a binary bit,
with the set of all neurons' rates encoding a bit sequence or 'pattern',
and when individual synaptic weights are tuned such that stored patterns
correspond to stable dynamical attractors. Most existing models use
morphologically degenerate neurons of one or two isopotential compartments
and suffer from poor stability and robustness. To study the effect
of cell morphology on network function, two networks were constructed
with identical connectivity, one composed of two-compartment neurons,
and the other of morphologically accurate 3D multicompartment models
of mouse layer 2/3 neocortical pyramidal neurons reconstructed from
confocal image stacks. Spatial distributions of ion channels and synapses
were tuned such that both cell models were equally excitable in response
to synaptic input. To evaluate the networks' functional properties,
an input was applied until the network's pattern reached a steady
state, then removed for a 'mnemonic period' during which the pattern
was maintained, and finally a second, different input was applied.
Stability and robustness, the abilities to maintain a pattern in the
presence of dynamical or structural perturbations, respectively, were
evaluated. Stability was quantified as the minimum strength of the
second input current required to alter the network's pattern. Robustness
was measured as the minimum variation of network-wide synapse strength
required to eliminate the network's ability to maintain a pattern
through the mnemonic period. Improvements in the stability and robustness
of the network of 3D neurons relative to the 2-compartment network
were measured, demonstrating the impact of realistic cell morphology
on models of the activity of highly interconnected neuron populations.
Support Contributed By: NIH grants DC05669, MH58911, RR16754, AG00001
|
 |
|
|
|
