Results of the proposed approach for synthetic single-spike-evoked transients with fast rise times.

<p>(<b>A</b>): QGIF model, (<b>B</b>): FHN model with high SNR, (<b>C):</b> FHN model with low SNR. (<b>A</b>-<b>C</b>) Inferring neuronal dynamics from transients with fast rise times (e.g. 3 <i>ms</i>). The fluorescence traces (first row) and [Ca²⁺] kinetics (second row) were simulated by using (<b>A</b>) QGIF and (<b>B</b> and <b>C</b>) FHN generative models. Inverting each model for its corresponding trace by using CaBBI led to a reasonably accurate inference about the membrane potentials and non-saturating [Ca²⁺] kinetics (rows four and three); 95% confidence intervals are shown as shaded areas. The inferred spikes (fourth row) coincide accurately with the veridical spikes (grey stars). The robustness of the method against high levels of noise is evident in the results shown in <b>C</b>, where only one spike was missed (red star). Fifth row shows the zoom into the rise time of a representative fluorescence transient in each trace shown in the first row. Membrane potential has the unit of [<i>mV</i>] in the (family of) QGIF model(s), but is dimensionless in the FHN model. The fluorescence traces were sampled at a frequency of 33.3 <i>Hz</i>. The values of parameters for the simulations and inversions can be found in <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004736#pcbi.1004736.t001" target="_blank">Table 1</a>, and prior distributions in <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004736#pcbi.1004736.t002" target="_blank">Table 2</a>. Units and conventions, as well as the parameter values are consistent across the following figures, unless stated otherwise.