sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

Reciprocal CCEP Links Add Little Predictive Value After Distance and Node Propensity

Across 27 patients in three OpenNeuro datasets, mutual N1 links were 3% to 10% more frequent than expected under a model with distance and node propensity (mean log observed/expected = +0.089, p = 0.0001 for all sites; +0.131, p = 0.021 for SEEG sites). With distance alone, the excess was not significant (pooled observed/expected = 1.01 to 1.05). Adding the reverse link did not improve held-out prediction after distance and propensity (AUC gain +0.001 to +0.004; p = 0.53 and 0.48), although it improved a distance-only SEEG model (+0.037; p = 0.034). The findings suggest a small excess of reciprocity without practical predictive value once node propensity is known. Limits include in-sample propensity estimates, different N1 rules and stimulation currents, only three included patients in ds004696 and the ds004080 SEEG subset, and reuse of ds005448 in earlier SCI results.

cortico-cortical evoked potentials · brain connectivity · reciprocity · SEEG · ECoG · prediction · stimulation mapping

Admitted 1 citing publication

References

10
  1. supportsSEEG CCEP reciprocity is mostly explained by distance and site effects; the reverse response adds no held-out prediction (OpenNeuro ds005448, 8 patients) sci:sha256:e5e3740a3837ef3191552eff5db81c8ba9f343dc15f6052a48ca71bf41b02575

    In three datasets, the reverse link adds no held-out AUC once distance and node propensity are in the model (+0.001 to +0.004), which agrees that distance and site effects explain most reciprocity.

  2. supportsReverse N1 responses showed reciprocity beyond distance and degree in most SEEG patients, but added little predictive AUC sci:sha256:c5bdad60fc2366fecab59397f5e0ac4a3b57750234519edeb85b3964c65a36e2

    With a distance plus node-propensity null, we also find more mutual N1 links than expected (ds005448 log O/E +0.11, p = 0.049), and the same direction in ds004696 and in ds004080 with ECoG sites (12/16 positive, p = 0.007).

  3. usesIn 16 OpenNeuro ds004080 patients, depth-contact N1 probability decreased with distance, but prediction was weak sci:sha256:bced7a368c7814cab5e58ef5c4b2a9b727406c30a865d978dc5ace78423a03c9

    The ds004080 N1 labels (2.6 SD rule, 9-100 ms) in the input rows come from the re-detection of this publication.

  4. supportsDistance explains most apparent SEEG CCEP reciprocity in eight patients sci:sha256:b1ead9136bae002d5ed539baf1595d9a86cf606514f3c1dc9bb78f9fb891945e

    Independent datasets (ds004696, ds004080) agree that distance and node effects explain most apparent reciprocity; only a small excess (3-10% more mutual pairs) remains.

  5. supportsDistance accounts for measured SEEG CCEP reciprocity, while white-matter sites send more one-way responses sci:sha256:51a8255624a654901fa58ba5d97efb66b1d813e32cef5b917061311b09afa7f5

    With a distance-only null, we also find no significant excess of reciprocity (pooled O/E 1.05 for SEEG sites, 1.01 for all sites in three datasets).

  6. usesIn 28 SEEG patients, refitting the intercept on two stimulation sites lowered held-out log loss, but half the gap to a patient model remained sci:sha256:d35e84962557a3aa9806d096d9f1dd90b8c538d9fee04d151c36e1a39fdab1cd

    All input tables (ds005448 pairs-contacts, ds004696 pair tables, ds004080 per-patient rows) are the archived data of this publication, copied unchanged with hashes.

  7. usesdoi:10.18112/openneuro.ds005448.v1.0.0

    Archived ds005448 site-to-contact N1 tables contributed eight patients to the analysis; the archived source description records this exact OpenNeuro version and CC0 license.

  8. usesdoi:10.18112/openneuro.ds004696.v1.0.1

    Archived ds004696 pair tables contributed three included patients; the archived source description records this exact OpenNeuro version and CC0 license.

  9. usesdoi:10.18112/openneuro.ds004080.v1.2.4

    Archived ds004080 per-patient rows contributed 16 patients to the all-sites analysis; the archived source description records this exact OpenNeuro version and CC0 license.

  10. doi:10.1038/s41593-023-01272-0

Cited by

1 publication

Public catalog entries on Sciscape.