SPIDER stitches brain recordings to reveal whole-brain information flow

A major challenge in neuroscience is mapping effective connectivity — the directed flow of information between brain regions — when recordings are fragmented across sessions, animals, and labs. Traditional methods like Granger causality require simultaneous recording with a shared clock, which is rarely possible for whole-brain coverage. Now, a new framework called SPIDER (Stitched Power-spectra for Inferring Directed information flow) solves this by using a non-parametric, frequency-domain approach. It stitches local power-spectral estimates from overlapping channel subsets into a global spectral matrix, then applies canonical spectral factorization and partial directed coherence (PDC) to infer directed interactions. Nuclear-norm completion fills in missing connections between regions never co-observed, all without temporal alignment.

SPIDER was rigorously validated on simulations, two-photon calcium imaging, and the International Brain Laboratory’s Neuropixels dataset. From 43 sessions across 12 laboratories — covering 50 brain areas never recorded together — it recovered brain-wide spontaneous flow. The results showed largely recurrent activity but a striking feedforward hierarchy in the theta band originating from the hippocampal formation. This finding was replicated in human resting-state intracranial EEG from 43 patients with non-overlapping coverage, demonstrating cross-species and cross-modality consistency.

The work makes whole-brain effective-connectivity analysis tractable for multi-session, multi-animal datasets that were previously incompatible with directed-flow inference. By enabling researchers to combine data from dozens of disparate recordings, SPIDER opens the door to large-scale collaborative neuroscience without requiring a single, synchronized recording.