Input segments without a nucleus reduced orientation prediction error in fixed tests on 3,461 MICrONS neurons
We tested how source coverage changes prediction of preferred visual orientation in one MICrONS mouse cortex volume. We used 3,461 neurons with manual registration and reliable model-derived orientation estimates. A fixed kernel model used soma position, visual area and cell class. Five outer folds withheld preferred-angle labels. Three inner folds chose model settings. The initial graph kept 81,792 neurons with one identified nucleus. Adding shared inputs reduced squared circular-coordinate error by 0.46% in random neuron folds. The gain was 0.26% in spatial folds, with a descriptive 95% block range of -0.01% to 0.54%. We then added source segments without a nucleus in the volume. This exploratory graph had 10,228,011 input edges and 11,615,820 synapses. It reduced error by 3.17% against the position and class reference in neuron folds and 2.34% in spatial folds. A further control held input totals and kernel diagonals the same. Adding shared inputs from segments without a nucleus then reduced error by 2.14% in neuron folds and 2.68% in spatial folds. The corresponding conditional block ranges were 1.35% to 2.87% and 2.08% to 3.25%. Thus, a nucleus requirement discarded useful prediction information under these fixed models. The later checks were chosen after the initial results. Segments can be incomplete axons or segmentation fragments. The data do not establish synaptic causation or transfer to new mice. The package preserves code, numerical inputs, source hashes, all predictions and fitted states, figures, and repeat instructions.
MICrONS · connectome · visual cortex · orientation tuning · shared input · source coverage · kernel regression · spatial validation
References
6- usesdoi:10.1038/s41586-025-08790-w
We use the MICrONS minnie65 v1507 static connection, nucleus and registration exports.
- usesdoi:10.1038/s41586-025-08840-3
We use exported orientation estimates and build on its analysis of common inputs and visual response similarity.
- usesdoi:10.1038/s41586-025-08829-y
We use the digital twin properties fitted from MICrONS functional recordings.
- usesdoi:10.1038/s41586-024-07765-7
We use cell class annotations that include the published soma and nucleus classification method.
- Adding a count of paths within one map reduced Brier error by 0.000014 in a fixed test of 21,074 C. elegans pairs without recorded direct edges sci:sha256:b2b1344ccd807499f1999100e88608ba09fd0e812aa20c6fa9dd3d4c91c18c6e
Related graph coverage work keeps structural worm relays without usable functional endpoint records. Our mouse test measures prediction information from input source segments without a nucleus.
- doi:10.1016/j.cell.2025.05.007
Prior fly work tests the limits of predicting neural response properties from synaptic inputs.
Cited by
9 publicationsPublic catalog entries on Sciscape.
- usesRequiring valid axon review marks reduced prediction gains for estimated orientation from 2.23% to 0.18% in a fixed kernel test of 3,461 neurons from one MICrONS mouse
We reuse its fixed target cohort and source coverage result.
- usesIncluding contacts from the same source changed the sign of an orientation similarity score on 354 MICrONS targets with tuning assigned by source
Its source selection and data provenance helped define this new spatial input extract.
- Under a binary source sign model, unknown signs left the covariance of inhibitory input fraction with model motion selectivity undetermined in 8,061 MICrONS neurons
Its source coverage tests motivate retaining input fragments.
- usesShared input identities reduced estimated orientation prediction error by 2.23% after position controls in a fixed test of 3,461 neurons from one MICrONS mouse
We use its fixed neuron cohort and compare the complete selected input count matrix with its earlier extract. We add input positions and local source label controls.
- usesShared output targets gave no clear extra gain in a fixed ridge test of estimated orientation in 3,461 neurons from one MICrONS mouse
We keep its fixed 3,461-neuron cohort and compare our new output extract with its earlier nucleus output matrix.
- usesSource groups gave opposite model orientation contrasts under 25 micrometre contact shuffles in one MICrONS mouse
We reuse its target cohort and attributed target selection code. We add local contact references. Our association score differs from its prediction test.
- usesCleft size weights reduced model orientation prediction error by 1.36% after count and position controls in 3,461 selected neurons from one MICrONS mouse
We reuse its fixed 3,461-neuron cohort and complete eligible input count graph, including sources without nuclei.
- usesA penalty control reduced the shared input gain from 1.49% to 0.69% in a fixed test of estimated motion polarity in 1,721 MICrONS neurons
We reuse its exact selected cohort, expanded input counts, source versions and source rebuild code. We add motion properties, a conditional prediction method and new controls.
- In one V1DD mouse, shared inputs from imaged cells had no clear link to movie noise correlation in 6,416 pairs after controls
Its MICrONS source-coverage result motivates a coverage audit for measured trial responses in V1DD.