Detection of synaptic inputs in model neurons with current-based or conductance-based synaptic inputs.

<p>A) Membrane potential traces recorded in a neuron <i>in vitro</i> (“Observed”) and simulated traces for two adaptive-exponential integrate-and-fire neuron models receiving either current-based (Model-I) or conductance-based (Model-g) synaptic inputs. In all three cases, input was generated using the same set of presynaptic spike times and synaptic weights. After optimizing model parameters, both models accurately reproduce the observed voltage fluctuations and spike timing of the recorded neuron. B) Detectability of synaptic inputs of different strength in the two models. Log likelihood ratio between models with no coupling and with constant-gain coupling plotted against input amplitude, for firing rates of at ∼1Hz, ∼5Hz, and ∼10Hz, as indicated. To illustrate the more reliable detection of excitatory inputs, lighter colors show the average log-likelihood ratio curves reflected about input amplitude 0. C) Correlation between the estimated coupling coefficient and true PSC amplitudes (5 Hz data). Dashed lines denote linear fits. D) Time necessary for input detection against its amplitude, for firing rates of at ∼1Hz, ∼5Hz, and ∼10Hz. Solid curves denote fits for the function c/x². Note that detectability of inputs in the models with current-based and conductance-based synapses are similar for two models and to the detectability of inputs in recorded neocortical neurons (compare to <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004167#pcbi.1004167.g004" target="_blank">Fig. 4</a>). Performing the same pair-wise analyses as above we find that, despite difference in the form of the input, once spike statistics are matched, the detectability of inputs of different strengths (A), estimation accuracy for the 5Hz condition (B), and detection times (C) are comparable to the experimental data. As before filled dots in (B) denote cases where Model 2 with coupling, provided significantly better fits than Model 1 with no coupling (Chi-square test, p<0.05). Error bars denote SEM. Dashed lines in (C) denote linear fits. Solid curves in (D) denote fits for the function c/x².