Statistical inference of changes in input strength.

<p>A) Example traces of aEPSCs before and after an amplitude change. In experiments, these traces were immersed in fluctuating noise current. B) Model fits of a data set consisting of two recordings from the same cell but with different amplitudes of aEPSCs (1.5σ and 0.5σ). Model 2 (M2, Constant Coupling), captures the effect of postsynaptic spike history and coupling to a presynaptic input on the full data. However, to detect changes in input strength these parameters must change over time. In Model 3 (M3, Variable Coupling), we separate the coupling effect into one regime before the change-point and one after the change-point. The post-spike history is the same before and after. C) Auto- and cross-correlations for the full data (left), and cross-correlations for the pre-change and post-change portions. Note that while Model 2 and Model 3 both describe the full data well, only Model 3 with variable-gain coupling describes separated pre-change and post- change portions. For comparison, the horizontal gray line denotes the peak of the cross-correlation when the full data is used. D) Log likelihood ratio between Models 2 and 3 for various amplitude changes (200s of data for all cells/changes). Black dots denote cases where Model 3, with variable-gain coupling, provided significantly better fits than Model 2, with constant-gain coupling (Chi-square test, p<0.05). E) Recording time necessary to detect changes in input amplitude. The dashed curve shows the detection times for constant-amplitude inputs (from <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004167#pcbi.1004167.g002" target="_blank">Fig. 2G</a>). Note that only slightly more data is necessary to detect changes in amplitude than to detect an input. Gray dots denote recording length for those cells where no change was detected. If these inputs are detectable the detection time must be longer than the recording length, as indicated by the arrows.