Fig. 4
From: Inferring and validating mechanistic models of neural microcircuits based on spike-train data
Estimation results for artificial synaptic input perturbations using in vitro data. a Log-likelihood ratio (LLR) of the I&F model with/without coupling on test data (10-fold cross-validation) as a function of recording duration for fixed aEPSC strength (in unit of σI, the standard deviation of the fluctuating input current) using methods 1a (left) and 2 (right). Shaded areas in a and b denote mean ± standard deviation across all cells. LLR > 0 indicates successful detection. b CCG peak z-score as a function of recording duration for aEPSC strength 1 σI. Dashed line indicates the significance threshold (95th percentile of the standard normal distribution). c Detection time for each cell as a function of aEPSC strength (in unit of σI) for method 1a (left), 2 (center), and the CCG method (right). Lines connect detection results for the same cell, light dots denote misses. In those cases, the detection time, if it exists, must be longer than the recording length, as indicated by arrows. * significant reduction of detection time (p = 0.0276, two-tailed paired t test, sample size n = 32). d Estimated perturbation strength z-score using method 1a (top) and CCG peak z-score (bottom), both as a function of aEPSC strength for 200 s recording length. Recordings were split into 200 s segments with up to 40 s overlap. Dashed lines indicate the significance threshold (cf. b). Dark dots mark successful detections
