Fig. 5

Nonlinear amplification in assemblies (A) Mean firing rate of assembly E neurons or the corresponding pseudo-assembly neurons as a function of mean afferent input to (pseudo-)assemblies (summed firing rates of all connected mitral cells). Data are from one rand and its corresponding struct network (n = 20 odors, 20 [pseudo]-assemblies). Lines: exponential fits; vertical dotted line: threshold for ?activation? in (D). (B) Gain index of assemblies (mean activity during vPIN normalized by Ctrl) as a function of input strength (same networks as in A). Each datapoint represents an assembly-odor pair (rand not shown for clarity), and the lines are linear fits (rand: FFI: r = ?0.34, p < 0.0001: FBI: r = ?0.25, p < 0.0001; struct: FFI: r = ?0.18, p = 0.0003; FBI: r = 0.11, p = 0.023). (C) Gain index as a function of the mean Dp firing rate during vPIN. Each datapoint corresponds to the gain index in response to one odor averaged over 20 struct networks. Lines: linear fits (rand: FFI: r = ?0.23, p = 0.33; FBI: r = 0.46, p = 0.04; struct: FFI: r = ?0.22, p = 0.35; FBI: r = 0.72, p = 0.0003). (D) Cumulative frequency of shared assembly activation in odor pair-network combinations that exhibited runaway correlations during vPINFBI (?r > 0.25; dashed line) or not (solid line). An assembly was defined as ?activated? when the total afferent input exceeded 240 Hz (vertical line in A).