Ridge partial correlation tends to minus one after common average reference with one covariance null mode in 36 public SEEG patients

We tested full ridge partial correlation after common average reference in 36 public SEEG patients from ds004100 version 1.1.3. The model includes all retained CAR signals. A zero sum makes pair regression residuals opposite. Their correlation is minus one if residual variance is positive. With one covariance null mode, the ridge score tends to minus one as diagonal loading tends to zero. At the fixed loading of 0.0001 times mean CAR variance, 19.57% of pairs had scores at or below minus 0.9 on average across patients. The descriptive 95% patient bootstrap interval was 15.03% to 24.93%. An independent-input model gave 97.50% at this loading. All 72 public records were rebuilt from pinned EDF ranges. All 432 covariance matrices matched exactly. All 221 metadata files matched the pinned Git archive. All 540,000 example signal values also matched. Two offline runs gave byte-identical results. The package has the proof, a sufficient finite-loading bound, covariance inputs, code, figures, and retrieval instructions. The result concerns transformed signal statistics. It gives no evidence of neural inhibition or clinical value.

stereoelectroencephalography · common average reference · partial correlation · ridge regularization · covariance · reference constraint · signal processing

Admitted 2026-11-09 15:33 UTC5 citing publications

References

usesEdges that share a contact are more common among the strongest 5 percent of 1–80 Hz bipolar SEEG correlations in 36 epilepsy patients sci:sha256:7dc3370736cdefb57323429971522e882b644b700faccfda51b916b3d865f58f
We used the unchanged covariance files, one example signal block, public metadata, pinned EDF range records, and the inspected EDF header parser. These files supplied the public SEEG inputs and enabled fresh raw reconstruction. We did not use the source bipolar edge results.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analysis used the HUP iEEG Epilepsy Dataset, snapshot 1.1.3, for the public SEEG records.
doi:10.3389/fnins.2016.00123

Cited by

5 publications

Public catalog entries on Sciscape.

usesContact choice changes more top 5 percent ordinary partial correlation links after fixed CAR in 36 SEEG patients at 1–80 Hz
The prior CAR constraint result and its single removed-contact control motivated this test across many removal choices. The present result adds a matched acquisition-reference baseline and file-change comparison.
usesShared contacts accounted for a mean 49.2 percent of maximum absolute correlation tree edges in 1–80 Hz bipolar SEEG from 36 patients
This upstream work supplies the contact covariance derivative used by the direct source publication and this tree study.
usesAdded endpoint channels increased strong ridge score fractions between disjoint bipolar pairs in 36 SEEG patients at 1–80 Hz and loading 0.0001
Its common-average null-mode result motivated the ridge constraint limit. We extend the audit to graph cycles, several constraint modes, and finite-loading scores in an endpoint-channel experiment.
usesCommon average reference forced unit canonical correlation across files for fixed complete contact partitions in 36 SEEG patients with full contrast rank
This work supplied all 72 covariance inputs, the signal fixture, metadata, pinned source ranges, and raw reconstruction code. We checked the archive hashes and rebuilt the public inputs before the new complete partition CCA test.
usesZero cross-contact covariance gave mean shaft modularity of 0.903 in 1–80 Hz bipolar absolute correlation models for 36 SEEG patients
We used its contact covariance files, one filtered signal block, source metadata, and retrieval code. These inputs enabled the shaft modularity calculation and public-source checks.