Estimator quality and anatomical matching constrain cross-cohort neural manifold flow inference: a case study in two IBL Neuropixels datasets

Abstract

We apply a new mathematical framework - Noetic Diffusion Theory (NDT) - to two large publicly available rodent electrophysiology datasets from the International Brain Laboratory, and attempt to test whether a discovery effect survives a structured external portability audit. An apparent association-cortex finding in the IBL Brain-Wide Map (BWM) - the Jacobian spectral radius predicts reaction time - does not survive the documented residual-quality gate: only 3 of 23 association-cortex sessions pass, so a quality-filtered correlation cannot be estimated. Applying the same pipeline to IBL Reproducible Ephys revealed a deeper problem: the two cohorts record from anatomically distinct cortical areas under the same broad region label. BWM "association cortex" is mainly retrosplenial and posterior parietal cortex; Repro Ephys association cortex" is 60% higher visual cortex (VISa/VISam), the primary target of the RepeatedSite probe. A pre-specified RSP-only replication gate confirmed that Repro Ephys contains zero RSP sessions and cannot serve as a replication cohort for the RSP/association-cortex claim. This paper is a case study showing how region-level NDT effects can fail under estimator-quality control and become non-portable across cohorts when broad atlas labels hide anatomical mismatch - a cautionary finding for any cross-dataset Neuropixels analysis.