A 7 Hz causal filter reader reduced unseen input squared error by a mean 50.4 percent at 50 ms in nine SEEG rest records

We tested forecasts of a specified causal filter in nine public SEEG rest records. The filter center was 7 Hz. Its decay time was 100 ms. We moved the known filter state forward by 50 ms. We then trained a linear reader to predict the added sum from unseen voltage. The mean squared-error reduction was 50.41 percent beyond known state. The descriptive 95 percent record bootstrap interval was 39.41 to 60.85 percent. All nine record means were positive. Independent noise gave minus 0.24 percent added skill, although known state explained 36.66 percent of future state energy. A secondary held-voltage forecast gave 35.19 percent skill. The reader had 1.37 degrees less mean phase error than this simple forecast. This is an exploratory test on reused clinical records and selected contact pairs. The target is a numerical filter state. Biological phase and stimulation effects remain untested. The package includes voltage, source versions and hashes, code, fitted readers, scores, three figures, proof, and reproduction commands.

causal filter · SEEG · signal prediction · state estimation · squared-error skill · ridge regression

Admitted 2026-11-09 19:00 UTC1 citing publication

References

usesRefitted spectrum controls gave 20.82 percent forecast skill at 50 ms for a future theta filter term in seven SEEG rest records sci:sha256:5331dccbbbce102efe12f29407e02b3847aa59bd5e06300de1073bc5d50f7e3b
Its spectrum controls guided interpretation of positive linear prediction skill. We make no claim that the new causal-state result exceeds what recorded temporal covariance can explain.
usesA 500 ms past voltage sum gave a 63.26 degree error for centered theta phase 50 ms ahead in seven SEEG rest records sci:sha256:678ac0e1b4e8d00b5e72eb22a8e9bbbfc9c4bd67c224a77925cb4cba677fc26f
This work supplied the known-versus-unseen filter contribution score concept. We applied the split to a causal state target that stops at the forecast time and used an exact state propagation comparator.
usesA shared reader had 58.38 degrees of mean absolute phase error in two new SEEG rest records at 50 ms with a centered theta reference sci:sha256:6bd2f301da5cc69ca85251358ec5d79565bf3c48c78038666bfdd36352965b56
This archive supplied the nine native-rate voltage extracts, their hashes, and source timing and contact checks. We used all nine extracts without new pair selection.
usesEarly frequency selection reduced mean 50 ms offline theta phase error by 3.55 degrees in seven SEEG rest recordings sci:sha256:5c3ae075960de0774e0fcc176ee07a5e3c9a353d86238f1e49023a23aa4d2bfa
This work supplied the original seven-record selection and source extraction evidence. We retained its selected pairs and pinned source metadata.
usesdoi:10.18112/openneuro.ds006065.v1.0.0
The study used the selected voltage extracts from OpenNeuro ds006065 version 1.0.0.
usesdoi:10.18112/openneuro.ds005670.v1.0.0
The study used the selected voltage extracts from OpenNeuro ds005670 version 1.0.0.
doi:10.1038/s41467-025-59417-7

Cited by

1 publication

Public catalog entries on Sciscape.

supersedesA 7 Hz causal filter reader reduced unseen input squared error by a mean 50.4 percent at 50 ms in nine SEEG rest records
This package corrects the archive file check. The check now verifies recorded files and permits added local environment files and sci retrieval metadata. All data, reader code, scores, and figures have the same hashes as the first archive.