Negative edges and soft thresholding in complex network analysis of resting state functional connectivity data

摘要: Complex network analyses of functional connectivity have consistently revealed non-random (modular, small-world, scale-free-like) behavior hard-thresholded networks constructed from the right-tail similarity histogram. In present study we determined properties resulting edges constrained to specific ranges across full correlation histogram, in particular left (negative-most) tail, and their dependence on confound signal removal strategy employed. absence global correction, left-tail comprised predominantly long range connections associated with weak correlations were characterized by substantially reduced modularity clustering, negative assortativity γ < 1 Deconvolution signals (white matter, CSF motion) resulted most robust within-subject reproducibility parameters (ICCs~0.5). Global altered topology clustering coefficient converging zero. Networks absolute value thus compromised following since different topologies mixed. These findings informed construction soft-thresholded networks, replacing hard thresholding or binarization operation a continuous mapping all values edge weights, suppressing rather than removing weaker avoiding issues related fragmentation. A power law adjacency function β = 12 yielded modular whose agreed well corresponding values, that reproducible repeated sessions many months evidenced small-world-like scale-free-like properties.