First, we counted the occurrences of each possible electrical triplet pattern (Figure 4A). The recorded quadruplets were separated

into triplets for a total of n = 173 triplets. The intersomatic distances measured for each configuration were used to predict the probability of electrical and chemical connections for the nonuniform random model. The occurrences predicted by both random models were counted in the same way as for the data (Supplemental Experimental Procedures). The ratio (data/prediction) indicates the relative occurrence of each of the four possible nonisomorphic patterns, Selisistat ic50 compared to the two random connectivity predictions (Figure 4A). We found that the predictions of both random BLU9931 connectivity models differ from the data. The uniform random prediction shows large deviations compared to the data for most patterns (p values: p1 = 0.003, p2 = 0.022, p3 = 0.0004, p4 = 0.0004), confirming that the model is insufficient to describe the statistics of connections of the MLI network. The nonuniform random prediction also deviates from the data but to a lesser degree, as the occurrence of fully connected triplets (pattern 4) is correctly predicted (p values: p1 = 0.0004, p2 = 0.213, p3 = 0.0004, p4 = 0.202). We separately confirmed that the fully interconnected triplets

(pattern 4) are indeed the result of direct connections and not indirect electrical coupling (Figure S4E). To characterize the electrical connectivity with a single measure and compare it to random connectivity models, we used the clustering coefficient C. C was originally introduced as a measure of the topological organization of networks and used to tuclazepam highlight differences between small-world networks and random networks, whose average C are significantly different ( Watts and Strogatz, 1998). C is usually measured for each node

in a network. Here, we calculate C for the recorded subnetworks of triplets and quadruplets of MLIs and compute the average over the configurations where C could be measured ( Supplemental Experimental Procedures). It should be noted that the average C obtained in this way is not intended to represent the average C of the whole network but is used to compare with C predicted by random connectivity models, where it was also calculated for subnetworks of triplets and quadruplets. For triplets, C effectively measures the likelihood that if neurons A and B, and B and C are connected, then A and C are also connected. The nonuniform random model predicted a higher clustering coefficient for electrical synapses, CE, than the uniform random model. This is expected if the tested neurons are sampled locally, as they were in the experiments ( Figures S2B and S2C). However, CE of the data significantly exceeds even the nonuniform random prediction ( Figure 4B; uniform random p = 0.0001; nonuniform random p = 0.0001).