Abstracts

Rationale: A neural network simulation using a model with realistic cortical architecture has been previously used to study synchronized bursting dynamics as a seizure representation (1). This model has the interesting property that bursting epochs arise spontaneously and cease at random intervals, making the onset of the seizure-like episodes difficult to predict. Recently, a point process model utilizing single unit recordings from human cortex has been demonstrated to show single neuron spike prediction abilities (2). We now test a similar point process prediction model on our neocortical simulation using random background inputs mimicking the role of surrounding cortex, with the goal of characterizing the subnetworks which are active in the generation of epileptogenic behavior. Methods: Point process models were used to fit the spiking probabilities in 10-msec time bins for a neural network simulation of neocortex. This simulation includes several classes of excitatory and inhibitory single-compartment neurons arranged and connected in space in a biophysically realistic fashion (1). The fitting technique was similar to that used by Truccolo et al. (2). Fitting terms corresponding to a given neuron s intrinsic spiking history as well as the extrinsic spiking history of randomly selected neighboring neurons were included. The fitting parameters were determined using the method of maximum likelihood conditioned on the observed spiking history over preceding time epochs of 100-msec in duration. Results: The point process model has been tested on 10 seconds of spiking data from pyramidal cells in layer II/III of the neural network simulation. Figure 1 displays four pre-spike filter functions used in the point process model. They display large amplitudes at short time scales and taper off of at longer times. Figure 2 displays the resulting conditional probability of firing by conditioning solely on the intrinsic history for a single cell using ten pre-spike filters. Predicting spikes by thresholding over the conditional probability would not produce reliable results as evidenced by the discordance in timing between the peaks of the predictions (blue trace) and the actual spike times of the cell (red dots). Including the extrinsic history from a random subset of ~20 neurons in the vicinity of this cell improves performance. Conclusions: The point process model has been tested on spike trains from a neural network simulation of epilepsy (1). We find that conditioning solely upon intrinsic spike trains does not provide reliable prediction. We aim to use this model to further delineate in which instances one can reliably perform single unit spike prediction in a spiking model of epilepsy with random inputs. 1. Anderson WS, et al., Studies of stimulus parameters for seizure disruption using neural network simulations. Biol Cybern. 97 (2): 173- 194, (2007). 2. Truccolo W, et al., Collective dynamics in human and monkey sensorimotor cortex: predicting single neuron spikes. Nat Neuro. 13 (1): 105 - 111, (2010). Support: Charles H. Hood Foundation (FA, WSA), NIH-NINDS KNS066099A (WSA)