Extended Data Fig. 9: Robustness of eigencircuits.
(a) Robustness to measurement error. Maximum absolute correlation between each original eigencircuit and all eigencircuits from the connectome with Gaussian noise added. For complex eigenvectors, we regressed the real and imaginary components onto those of another then calculated the r-value of this two-parameter linear fit. Measure was taken for three levels of noise (SNR = 100 \({\sigma }^{2}=0.01\mu \) black, SNR = 10 \({\sigma }^{2}=0.1\mu \) grey, and SNR = 1 \({\sigma }^{2}=\mu \) red where \(\mu \) is the synapse count). (b) Non-anatomically localized eigencircuits tend to be less robust to measurement error. Robustness of eigencircuits measured as maximum absolute correlation with eigencircuits from noisy connectome (SNR = 1, see Extended Data Fig. 9 red trace) plotted against locality index, fraction of synapses in one neuropil for top 75% of eigenvector loading of each eigencircuit. (c) Robustness of eigencircuits to hyperbolic tangent transformation (tanh) of connectome weights scaled by half the max synapse count (c = 2). (d) Plot of maximum absolute correlation of original vs transformed connectomes (see A) versus rank of original eigencircuit, split across real and complex unique eigencircuits of top 250 eigencircuits. (e) Robustness of non-localized eigencircuits to choices in loading threshold. (left) Localization index computed across different inclusion criteria. (right) Number of neurons within each eigencircuit for multiple loading thresholds.
