In a positive projection model with equal leaks and gain 0.8, time scales had rank correlation −0.63 with measured values in 192 mouse regions when weights were input shares
We tested whether projections between mouse brain regions predict neural time scales. We used a positive linear model with equal leaks and independent unit noise. The full graph had 288 disjoint regions. The functional test used 192 regions, 73,560 units, and 423 sessions from an open time scale atlas. The recorded main test used each source region's output shares. It gave rank correlation -0.050 and adjusted partial correlation 0.008. An exploratory test used each target region's input shares. It gave rank correlation -0.632, with a 95 percent spatial resampling range of -0.738 to -0.469. The reverse relation remained at gains 0.5 and 0.95, with equal session weights, and in separate session groups. A negative regression slope let the input score reduce prediction error beyond anatomy controls. Across 20 region split seeds, median reductions were 6.31 percent with random folds and 4.38 percent with spatial folds. This statistical gain did not validate the model's physical time scale order. These results apply to the stated positive regional models and exported neuronal medians. They do not identify the role of local dynamics, inhibition, or shared noise. The package preserves new code, exact source hashes, atlas assignments, model scores, predictions, figures, and a full repeat with 20 exact numerical file matches. Original source data remain external under their source terms.
mouse · connectome · neural time scales · projection weights · structure and function · linear dynamics · weight normalization · reproducibility
References
10- usesdoi:10.5281/zenodo.21257745
We use the pinned publication archive exports of regional and per-unit neural time scale estimates. The original tables remain external.
- usesdoi:10.1038/s41586-025-09235-0
The time scale estimates derive from this IBL brain-wide recording resource. We use exported estimates rather than the raw spikes.
- usesdoi:10.1162/netn_a_00066
We use the regionalized voxel projection strength matrix, with a checksum-pinned server version dated 2025-02-21.
- usesdoi:10.1038/nature13186
This Allen tracer resource is the experimental basis of the regionalized projection model.
- usesdoi:10.1016/j.cell.2020.04.007
We use CCF 2017 region labels, volume, and position to build the atlas match and anatomy controls.
- doi:10.1016/j.neuron.2015.09.008
This prior circuit model explains why recurrence and local differences can affect regional time scales.
- At gain 0.5, visual path steps gave no clear timing prediction gain for 76 mouse regions in a 2024 IBL export sci:sha256:45e495a01b1372ed6878fd846d93dffbaa8e0db5815a6ceb04593ff7ed141776
This prior SCI study tested projection path steps against visual representation times. We test spontaneous neural time scales with new code.
- doi:10.1038/s41467-025-66542-w
This prior control model fits heterogeneous regional decay rates. It differs from our equal-leak model and functional target.
- doi:10.1101/2025.08.30.673281
This prior study defines multi-component neuronal time scales and reports brain-wide regional differences. The later functional source export uses this effective time scale definition.
- doi:10.48550/arXiv.2603.04149
This prior neuron-scale model links topology and time scale diversity. Our result is limited to regional positive projection models.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesWith input shares, equal leaks and gain 0.8, a positive mouse projection model had at least 3.05-fold time scale error in one of 192 regions for every independent white noise pattern
We reuse the atlas assignment, eligible region flags, published model scores, and separate session-half medians. We extend its equal-noise test to every independent white noise strength pattern.