With neuron input and output counts fixed, 45 three-cell loops were within the exact reference range in 168 human cortex recording graphs
We tested three-cell feedback loops in released human layer 2–3 recording graphs. The main panel had 168 complete epilepsy recording graphs, 1,132 neurons, and 1,123 directed links. We counted every binary graph with the recorded input and output counts at each neuron. The 45 recorded loops were within the exact central 95% reference range of 42–62. The reference mean was 51.92, and the lower-tail fraction was 0.1024. Fixing each graph’s reciprocal pair total gave a mean of 47.33 and a range of 39–56. A control that fixed only each graph’s link total gave a mean of 75.29. An added check found eight three-link loops with no reverse links. Their low reference tails depended on this stricter motif definition. This added result was exploratory. A linear model identity shows why extra reverse links do not remove the three-cell return term. The separate tumor cohort had six loops, within its exact range of 3–9. These are conditional tests of selected slice recordings. They do not show the absence of feedback or a lack of biological cell roles. The package contains source tables, graph masks, exact distributions, code, figures, and repeat instructions.
human cortex · local connectivity · feedback loops · neuron degrees · exact graph counts · conditional graph reference · induced motifs · linear response
References
6- usesdoi:10.1126/science.adg8828
We test a further loop question in its released local recording graphs.
- usesdoi:10.5061/dryad.t76hdr87b
We use the source tables and graph arrays from the 28 March 2024 release.
- usesdoi:10.1137/16M1087175
We use an established uniform graph reference with fixed neuron degrees.
- With chemical degrees and cell classes fixed, only three recordings could change link overlap in 8,527 selected mouse neuron pairs sci:sha256:9143ede6b3b5492a75d315b1f37d53d0903ce45bf14b31206f996ce7afe1cee0
It shows why small recording graphs can lose test variation after neuron counts are fixed.
- doi:10.1038/s41593-025-02134-7
This related study describes the source cell and synapse properties.
- usesdoi:10.6084/m9.figshare.28184126.v1
We checked its extraction code to confirm the meaning of source connection code 1.2. No source software is bundled.
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesPositive EPSP weights gave no clear loop product increase over shuffled weights in 151 human epilepsy resection panels
Its binary loop study and source audit motivated the weighted return test and the six-panel test mask check. We wrote new code and independently checked the source archive.
- usesOther outputs of the same neuron gave no clear strength prediction gain in shrinkage tests of 934 human layer 2–3 connections
We reuse its signed graph extract and source audit to test removal of mixed-component links and restriction to complete graphs.
- usesTwo-step route counts gave no clear strength prediction gain in a fixed ridge test of 995 human cortical connections
We reuse its MIT graph preparation code and its source test-mask audit.
- EPSP weights raised the mean fraction of stable gains that permit brief growth in 54 human cortical panel models with equal cell time constants
This earlier audit tests short cycle counts in the same public recording data. We cite it to distinguish the new weighted response question.
- Shared neighbors gave no clear prediction gain for response size in nested ridge tests of 903 mouse V1 connections
This related study motivates controls for neuron connection counts in small recording graphs.