Group Classification Algorithm
Our parameter variation resulted in
a parameter space of 10,485,760 model instances. We classified these simulated
HCO and isolated neuron instances by their activity characteristics, so
that instances showing the same electrical activity are segregated to the
same group.
The classification algorithm parsed
automatically the spike and voltage files corresponding to each simulated
instance and gathered information about its electrical activity. Based on
this information, the algorithm assigned a group label to each model
instance. The HCO instances were grouped into the following 10 groups: spiking, silent, asymmetric activity, plateau,
asymmetric bursting, irregular spikes, one burst, irregular period,
unbalanced, and functional.
The isolated neuron instances were split into the following 8 groups: spiking, silent, bistable,
plateau, irregular spikes, one burst, irregular period, and regular bursting. The type of electrical
activity characterizing each group can be visualized in Figures 2 and 4 of Doloc-Mihu
& Calabrese (2011).
Note: Our
classification method was designed such that there were no positive misses
(for example, we did not classify models with bursting
activity within the group of models with spiking
activity), but we could have false negatives (for example, we may classify models
with spiking activity within the group of
models with bursting activity).
We used eight activity characteristics
in the classification algorithm, as follows:
Step 1 (activity
type level):
1. Any spiking activity present or not
(threshold of -20mV) to obtain the silent and
non-silent HCOs
2. For non-silent HCOs, the number of
spikes – presence of minimum 3 spikes to define a burst or not
3. Presence of a minimum inter-burst
interval of 1 sec to differentiate between spiking
and bursting activities
Step 2 (spike
type level):
4. The value of the voltage trace at
the end point of the burst (greater than -35 mV) to obtain the plateaus
5. Coefficient of variation of the
amplitudes of the spikes within the burst (less or equal than 0.07) to
obtain the irregular spikes and the
normal spikes HCO instances
Step 3 (“number
of bursts” level):
6. For the later group, the number of
bursts per neuron in 100sec of simulation time to obtain the one burst instances and the repeated
bursts instances
Step 4 (period
regularity level):
7. For the repeated instances, the
coefficient of variation of the period (less than 0.05) to obtain the irregular period and the regular
period instances
Step 5 (phase
level):
8. For the regular period instances,
the mean phase (in the range of 0.45-0.55) to obtain the unbalanced and the functional instances.
The pseudocode
for the entire classification algorithm can be found in the Appendix 1 of Doloc-Mihu
& Calabrese (2011). Table 3 from Appendix 4 provides the statistics
for the HCO instances broken down by the synaptic components present.
Last updated June
22, 2012. Please send comments to adolocm@emory.edu.
|