Multi-step connectome communication measures add 0.4 to 1.9 percent of explained FC variance beyond direct wiring, distance and node effects in worm, fly, mouse, marmoset and human

Question: after we remove distance, direct wiring and node-level effects, do multi-step network communication measures of the connectome still predict functional connectivity (FC)? Is the answer the same across species? Approach: one pipeline for five reused data sets: worm (Cook 2019 connectome, Atanas 2023 calcium, 40 animals), fly (FlyWire v783 neuropil flow, Aimon 2023 light-field calcium, 50 flies), mouse (Knox 2019 tracer model, Grandjean 2020 fMRI, 79 mice), marmoset (Majka 2020 tracer, Schaeffer 2022 fMRI, 31 animals) and human (Lausanne DSI and fMRI, 69 persons). Pair models had node fixed effects, a distance spline, direct weight terms and a hub term. We added six measures: communicability, search information, shortest-path, diffusion and navigation efficiency, and navigation success. We used 10-fold cross-validation, split halves of individuals, and 100-500 surrogate graphs of up to three types (distance-bin shuffle, shuffle with node strengths kept, degree-preserving distance-matched rewiring). Result: the communication measures added a small gain in explained group FC variance in all five species: worm 0.0044, fly 0.0042, mouse 0.0081, marmoset 0.0191, human 0.0145 (cross-validated values within 0.001). Surrogate graphs gave mean gains 2.4-35 times smaller (largest p 0.027 in worm, 0.002-0.003 in the others; all pass a Holm correction). The gain was positive in every individual. It also occurred in pairs with no direct connection (worm, marmoset, human). The best single measure changed between species. In four sets, direct wiring added 6-27 times more variance than the communication measures; the gain is at most 5.6% of the remaining explainable FC variance. Limitations: FC is slow and correlational; connectome and activity come from different animals except in human; the marmoset graph has only 55 injected areas; rewiring nulls were possible only for the sparse worm and human graphs; the first rewiring null was too weak and was replaced (both results reported).

structure-function coupling · network communication · communicability · search information · navigation · functional connectivity · connectome · cross-species · C. elegans · Drosophila · mouse · marmoset · human · null models