Figure 1.
MEA and recorded signals overview.
(A) Dissociated cortical neurons coupled to a MEA. (B) Raster plot of the electrophysiological activity: each row corresponds to a recording site, and each small vertical line corresponds to a detected spike. (C) Electrophysiological activity recorded from one microelectrode.
Table 1.
Network model parameters considered in the simulations.
Figure 2.
Schematic overview of the considered connectivity methods.
(A) Binary string is created starting from the spike train. A window is selected to evaluate the TE and to define the MI symbols (window = 0.3 ms). (B) Cross-inter-spike-intervals (cISI) between neurons X and Y are highlighted by the red arrows. (C) Cross-correlation function between neuron 1 and 2. The directionality of the connection is evaluated considering the peak latency from zero. (D) Mutual information (spike count approach) function related to a pair of nodes of the network model. The inset shows that the MI peak value falls close to the zero time shift (value −0.1 ms).
Figure 3.
A small network consisting of five neurons is considered. (A) Network graph. The numbers indicated on the arrows are the synaptic weights. (B) SWM (left) and CM (right). The red, black and yellow circles on the TCM correspond to a true positive (TP), true negative (TN) and false positive (FP), respectively. (C) Positive Precision Curve. The red curve corresponds to the best performance a given method could achieve. The dashed black line corresponds to the number of synaptic weights present in the model (panel A). (D) By comparing the thresholded SWM (left) to the TCM (right) the blue curve (drawn as an example) of the positive precision plot (panel C) is obtained. The white background elements on the TCMs (right) correspond to the TFS elements being analyzed.
Figure 4.
ROC and PPC curves relative to the presented connectivity methods.
(A) ROC curves relative to the completely excitatory network models. (B) ROC curves relative to the network models where the connections are both excitatory and inhibitory. Black, blue, red, and green lines indicate CC, MI, JE and TE, respectively. The diagonal (dashed) line (A–B) corresponds to the random detection. (C) PPC evaluated on the E network models. (D) PPC evaluated on the EI network models. The insets (C–D) show a zoom of the first 100 TFS. The vertical dashed line (C–D) identifies the number of the excitatory elements present in the SWM.
Table 2.
AUC (Area Under Curve) values (mean±standard deviation) for the different connectivity methods.
Figure 5.
Evaluation of the AUC as a function of the bin size and of the temporal window.
(A–B) 3D and false color map representation of the AUC obtained by using the MI method. (C–D) 3D and false color map representation of the AUC obtained by using the TE method. (E) Bar plot of the AUC obtained by using the CC method.
Figure 6.
Connectivity maps obtained on experimental data (MEA) by (A–B) TE, (C–D) JE, (E–F) CC. The threshold values correspond to 40 and 200 links respectively.
Figure 7.
Overlap curves show the number of common links identified by different connectivity methods.
(A) Overlap curves from experimental data, (B) overlap curves from the synthetic dataset (mean±standard deviation). The dashed curves show the number of TP values commonly identified by both the methods. The inset highlights the first 600 connections.
Figure 8.
ISI parameters evaluated over simulated and experimental data.
(A) Fano Factor evaluated on the probabilities values (y-axis) of the ISI histogram. ISI bin size of 0.2 ms. (B) percentage of non-null ISI bins with respect to the total ISI number.