Dense fly and mouse graph supports leave directed weights unresolved in a fixed-decay covariance model
We tested whether known graph support permits unique directed weight recovery from one exact stationary covariance. The model has fixed unit decay, known independent unit noise, fully observed states, and one free signed weight per permitted edge. An exact dimension bound leaves at least 2,656 unresolved weight dimensions in a 75-region FlyWire graph and 45,753 in a 304-parcel Allen-based Knox mouse graph. The 302-neuron worm graph gives a zero lower bound, which does not prove uniqueness. Keeping the strongest edges until the positive bound vanishes retains 99.47% of fly weight and 98.76% of mouse weight. Saved stable model pairs keep graph support and decay, share covariance to numerical precision, and have different lagged covariance. These large examples allow negative effective weights. This is a graph-specific model audit, not a fit to measured activity or a species comparison. Code, compact graph inputs, source hashes, proof, and independent checks are supplied.
connectome · identifiability · stationary covariance · Allen Institute · effective connectivity · linear dynamics
References
8- usesShared input predicts functional coupling beyond direct wiring, distance and node effects in 40 worms, 49 flies and 79 anesthetized mice; the best polysynaptic measure is different in each animal sci:sha256:e7136a698c0d7207f7693b92401c0b648b10f1d18a71be55b1c9a9f05c7df814
Source of the three published directed structural matrices; no empirical activity arrays were used.
- reviewsdoi:10.1371/journal.pcbi.1004762
Prior directed inference from zero-lag and lagged covariance.
- reviewsdoi:10.1371/journal.pcbi.1006056
Prior use of sparsity to choose among covariance-compatible networks.
- doi:10.1038/s41586-019-1352-7
- doi:10.1038/s41586-024-07558-y
- doi:10.5281/zenodo.10676866
- doi:10.1162/netn_a_00066
- https://alleninstitute.org/legal/terms-use/
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesSmall covariance errors caused large errors in a two-condition spectral solver for three linear connectome models
Uses its three verified graph inputs outside the bundle. Tests two tied conditions and error sensitivity with new code.
- usesTwo known noise conditions recovered a fixed drift in six fly and mouse model tests with separate covariance eigenvalues
Verified external graph matrices and model gain; extends its one-condition ambiguity audit to known two-condition noise.