pith. sign in

arxiv: 2606.22695 · v1 · pith:UB7MCKQVnew · submitted 2026-06-21 · 🧬 q-bio.NC · stat.ME

SPIDER -- Stitched Power-spectra for Inferring Directed information flow from incomplete and asynchronous Experimental Recordings

Yisi S. Zhang , Daniel Y. Takahashi This is my paper

Pith reviewed 2026-06-26 09:13 UTC · model grok-4.3

classification 🧬 q-bio.NC stat.ME
keywords directed connectivityeffective connectivitypartial directed coherencepower spectral densitynuclear norm completionbrain networkstheta bandhippocampal formation
0
0 comments X

The pith

SPIDER recovers directed information flow among 50 brain areas from recordings never taken together.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents SPIDER, a method that stitches local power-spectral estimates from overlapping but incomplete channel subsets into a global matrix, uses nuclear-norm completion to fill never-co-observed pairs, and then applies canonical spectral factorization plus partial directed coherence to obtain frequency-resolved directed interactions. This works without any shared temporal reference across sessions, animals, or laboratories. A reader would care because standard directed-connectivity tools require every region to be recorded simultaneously on one clock, which rules out most existing large-scale datasets. SPIDER demonstrates the approach on simulations, calcium imaging, and the International Brain Laboratory Neuropixels data, then uses it to show that spontaneous brain-wide flow is mostly recurrent yet forms a clear feedforward hierarchy in the theta band with the hippocampal formation at the top, and recovers the same pattern in human intracranial EEG.

Core claim

SPIDER stitches local power-spectral estimates from overlapping channel subsets into a globally consistent spectral matrix, fills missing entries by nuclear-norm completion, and extracts frequency-resolved directed interactions by canonical spectral factorization followed by partial directed coherence, thereby recovering directed information flow among 50 areas from 43 sessions in 12 laboratories that were never recorded together and lack any common clock.

What carries the argument

Stitched global power-spectral matrix completed by nuclear-norm minimization, followed by canonical spectral factorization and partial directed coherence (PDC).

If this is right

  • Directed connectivity inference becomes possible for any collection of sessions whose region coverage overlaps sufficiently, even without temporal alignment.
  • Spontaneous activity across the brain is shown to be largely recurrent except in the theta band, where a feedforward hierarchy originates in the hippocampal formation.
  • The same theta-band hierarchy appears in both rodent Neuropixels data and human intracranial EEG, indicating consistency across species and modality.
  • Multi-session, multi-animal, multi-laboratory datasets that were previously unusable for effective-connectivity analysis now become tractable.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same stitching-plus-completion strategy could be tested on other frequency-domain network measures that also rely on a complete cross-spectral matrix.
  • If nuclear-norm completion succeeds here, analogous matrix-completion steps might allow directed-flow inference in other domains where only partial pairwise observations exist.
  • The method implicitly treats the power spectrum as the fundamental observable that can be assembled from fragments; direct validation against ground-truth simultaneous recordings at scale would strengthen that premise.

Load-bearing premise

Local power-spectral estimates from overlapping subsets can be stitched into a globally consistent matrix whose missing entries are recovered accurately enough by nuclear-norm completion that the subsequent factorization and PDC estimates remain undistorted.

What would settle it

Apply SPIDER to a dataset in which all 50 areas are recorded simultaneously, then compare the resulting PDC estimates against those obtained by standard PDC on the complete simultaneous recording; any systematic discrepancy would falsify the claim that stitching and completion preserve the directed-flow estimates.

Figures

Figures reproduced from arXiv: 2606.22695 by Daniel Y. Takahashi, Yisi S. Zhang.

Figure 1
Figure 1. Figure 1: Problem setting and pipeline overview. (A) Recording sessions cover overlapping but incomplete subsets of a network. (B) When PDC is estimated within a single session’s incomplete scope, a hidden common driver X3 induces a spurious directed connection between X1 and X2 that is indistinguishable from true coupling under local observation. (C) Stitched spectral matrix Sˆ(ω). Local cross-spectral estimates fr… view at source ↗
Figure 2
Figure 2. Figure 2: SPIDER recovers directed information flow under complete and incomplete observation and in bio￾physically realistic spiking networks. (A) Three-node VAR(1) system used for validation. X3 drives both X1 and X2, while X1 and X2 have no direct interaction. All three pairwise subsets are observed in independent asynchronous blocks. (B) PDC estimates (rows: target node; columns: source node) under complete pair… view at source ↗
Figure 3
Figure 3. Figure 3: SPIDER accuracy in high dimension. Simulations use K = 50-dimensional stationary VAR(1) processes (10 replicates). MSE of off-diagonal PDC entries is reported against (A) the theoretical PDC and (B) the PDC from a single fully simultaneous recording. Error bars are standard deviations across simulation replicates; asterisks denote pairwise significance from paired t-tests (∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < … view at source ↗
Figure 4
Figure 4. Figure 4: Directed information-flow reconstruction from mouse calcium imaging using SPIDER. (A) Estimation stability as a function of data length, quantified by mean squared error of PDC magnitudes and correlation of integrated information-flow matrices with the full-length estimates. Truncated datasets were generated via random circular shifts; points are medians and error bars indicate robust variability (1.4826 ×… view at source ↗
Figure 5
Figure 5. Figure 5: Directed information-flow reconstruction from IBL Neuropixels data. (A) Pipeline for area-level stitching from Neuropixels data. Spike trains from each brain area are reduced to a single canonical spectral mode via spectral PCA. Local cross-area spectra estimated from two partially overlapping sessions are assembled into the stitched spectral matrix and factorized via Wilson’s algorithm to calculate PDC. (… view at source ↗
Figure 6
Figure 6. Figure 6: SPIDER reveals a theta-band feedforward hierarchy in spontaneous brain-wide flow. (A) How hierarchy is quantified. A directed network is summarized by a trophic-incoherence index F0 (its “non-hierarchicalness”): each area receives a trophic level h, and F0 is the flow-weighted variance of the level difference across edges, equal to 0 for a perfectly feedforward hierarchy and approaching 1 when edge directi… view at source ↗
Figure 7
Figure 7. Figure 7: SPIDER recovers a theta-band feedforward hierarchy in resting human intracranial EEG. Resting in￾tracranial EEG from 43 patients (OpenNeuro ds003688), stitched over 38 left-hemisphere cortical AAL regions that were never recorded simultaneously. (A) Stitched directed information-flow (PDC) matrix; colour encodes integrated directed flow from source (column) to target (row), with the diagonal zeroed. (B) Tr… view at source ↗
read the original abstract

Mapping the directed flow of information between brain regions -- their effective connectivity -- is central to understanding brain function, yet large-scale recordings sample only a fraction of the brain at a time: sessions, animals, and laboratories cover different, partially overlapping regions, usually without a shared temporal reference. Established directed-connectivity methods (Granger causality, dynamic causal modeling, partial directed coherence, PDC) require all regions to be recorded simultaneously and with a common clock. We introduce SPIDER, a non-parametric, frequency-domain framework that recovers directed information flow from such incomplete, asynchronous recordings: it stitches local power-spectral estimates from overlapping channel subsets into a global spectral matrix and obtains frequency-resolved directed interactions by canonical spectral factorization and PDC, without temporal alignment, while nuclear-norm completion fills in never-co-observed region pairs. With consistency guarantees, we validate SPIDER on simulations, two-photon calcium imaging, and the International Brain Laboratory Neuropixels dataset, recovering directed flow among 50 areas from 43 sessions in 12 laboratories never recorded together. Beyond validation, SPIDER reveals what no single recording can: brain-wide spontaneous flow is largely recurrent, but in the theta band it forms a significant feedforward hierarchy with the hippocampal formation at its source. Applied to resting human intracranial EEG (43 patients, non-overlapping coverage), it recovers the same theta-band hierarchy across species and modality. SPIDER makes whole-brain effective-connectivity analysis tractable for multi-session, multi-animal datasets previously incompatible with directed-flow inference.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces SPIDER, a non-parametric frequency-domain framework for recovering directed information flow (via canonical spectral factorization and partial directed coherence) from incomplete, asynchronous multi-session recordings. Local power-spectral estimates from overlapping channel subsets are stitched into a global spectral matrix; nuclear-norm minimization completes entries for never-co-observed region pairs; the completed matrix is then factored to obtain PDC without requiring temporal alignment. The paper asserts consistency guarantees, validates the pipeline on simulations, two-photon calcium imaging, the IBL Neuropixels dataset (50 areas, 43 sessions, 12 laboratories), and human intracranial EEG (43 patients), and reports that brain-wide spontaneous flow is largely recurrent but exhibits a significant theta-band feedforward hierarchy with the hippocampal formation at its source, consistent across species and recording modality.

Significance. If the recovery claims hold, SPIDER would make whole-brain effective-connectivity analysis feasible for the growing class of multi-lab, multi-animal datasets that lack simultaneous coverage, directly addressing a practical barrier in systems neuroscience. The validation on both simulated data and two independent real-world datasets (IBL Neuropixels and human iEEG), together with the cross-species consistency of the theta-band hierarchy, would constitute a substantive methodological advance. The non-parametric construction and avoidance of explicit temporal synchronization are additional strengths.

major comments (2)
  1. [Abstract (stitching and completion steps)] Abstract (paragraph describing the stitching and completion steps): the central claim that nuclear-norm completion recovers a globally consistent Hermitian positive-semidefinite spectral matrix whose missing blocks do not materially distort the subsequent minimum-phase factorization or PDC estimates is load-bearing for all downstream results. No explicit guarantee, error bound, or validation of the regularization parameter is supplied to ensure the completed matrix remains positive definite or that phase information is preserved under the specific 12-lab, 43-session asynchronous coverage pattern; if this step introduces systematic phase distortion, the reported theta-band feedforward hierarchy with hippocampal source could be an artifact.
  2. [Abstract (consistency guarantees and validation paragraph)] Abstract (consistency guarantees and validation paragraph): the asserted consistency theorems are stated under idealized conditions that may not cover the multi-session, multi-laboratory regime actually used; the manuscript provides no derivation details, error-bar reporting, or description of how post-hoc choices (regularization, rank threshold) were validated on the target data, leaving the robustness of the 50-area recovery claims difficult to assess.
minor comments (1)
  1. The abstract is information-dense; separating the methodological pipeline description from the empirical claims would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential of SPIDER to enable directed connectivity analysis on multi-lab datasets. We address each major comment below with clarifications from the manuscript and indicate planned revisions.

read point-by-point responses

  1. Referee: Abstract (paragraph describing the stitching and completion steps): the central claim that nuclear-norm completion recovers a globally consistent Hermitian positive-semidefinite spectral matrix whose missing blocks do not materially distort the subsequent minimum-phase factorization or PDC estimates is load-bearing for all downstream results. No explicit guarantee, error bound, or validation of the regularization parameter is supplied to ensure the completed matrix remains positive definite or that phase information is preserved under the specific 12-lab, 43-session asynchronous coverage pattern; if this step introduces systematic phase distortion, the reported theta-band feedforward hierarchy with hippocampal source could be an artifact.

    Authors: The full manuscript (Methods and Supplementary Note 3) derives consistency of the stitched spectral matrix under sufficient pairwise overlap and shows via matrix completion theory that nuclear-norm minimization recovers the true Hermitian PSD matrix with high probability when the number of observed blocks satisfies standard incoherence conditions. The regularization parameter is chosen by cross-validation on simulations that replicate the exact 12-lab missingness pattern of the IBL data; these simulations also quantify phase distortion after minimum-phase factorization, which remains below 5° in the theta band for the observed completion error. The same theta-band hierarchy is recovered independently from the human iEEG dataset (different coverage, different modality), which would be unlikely if the result were driven by systematic phase artifacts in the Neuropixels completion step. We will add an explicit error-bound statement and a supplementary phase-error panel to the revised abstract and main text. revision: partial

  2. Referee: Abstract (consistency guarantees and validation paragraph): the asserted consistency theorems are stated under idealized conditions that may not cover the multi-session, multi-laboratory regime actually used; the manuscript provides no derivation details, error-bar reporting, or description of how post-hoc choices (regularization, rank threshold) were validated on the target data, leaving the robustness of the 50-area recovery claims difficult to assess.

    Authors: The consistency theorems (Theorem 1 and Corollary 2) are proved in Supplementary Note 2 under the precise multi-session asynchronous model used in the IBL analysis; the proof explicitly incorporates block-wise estimation with partial overlaps and the nuclear-norm step. Derivation details, including the overlap condition and the resulting rate, will be moved to the main Methods in revision. Error bars on simulation recovery metrics are already reported (Fig. 2); for real data we add bootstrap standard errors across sessions. Regularization and rank thresholds were selected by held-out session cross-validation on the IBL recordings (Methods, paragraph on hyperparameter selection); we will expand this paragraph with the exact validation procedure and sensitivity plots. revision: yes

Circularity Check

0 steps flagged

No circularity: SPIDER derivation relies on external validation and imposed modeling choices rather than self-referential fits

full rationale

The described pipeline stitches local power-spectral estimates from overlapping subsets into a global matrix, applies nuclear-norm minimization to recover never-co-observed blocks, performs canonical spectral factorization, and computes PDC. None of these steps reduce by the paper's own equations to quantities fitted from the target recordings themselves; the low-rank assumption is an explicit modeling choice, not a tautology, and the consistency guarantees plus validation on simulations, two-photon data, IBL Neuropixels (43 sessions, 12 labs), and human iEEG are presented as independent checks. No self-citation chain is invoked to force uniqueness of the factorization or hierarchy result, and the theta-band feedforward claim is an output of the processed data rather than a renaming or re-derivation of the inputs. This matches the expectation of a self-contained construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review; the central claim rests on the domain assumption that overlapping subsets permit consistent global spectral stitching and that nuclear-norm completion preserves directed-interaction structure. No free parameters or invented entities are explicitly named.

axioms (2)
  • domain assumption Local power-spectral estimates from overlapping channel subsets can be combined into a globally consistent spectral matrix
    Invoked in the description of the stitching step that enables the global PDC computation.
  • domain assumption Nuclear-norm completion accurately recovers never-co-observed region pairs without distorting frequency-resolved directed interactions
    Required for the claim that the method works on pairs never recorded together.

pith-pipeline@v0.9.1-grok · 5809 in / 1508 out tokens · 25513 ms · 2026-06-26T09:13:42.145917+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references

  1. [1]

    Dynamic causal modelling

    Friston KJ, Harrison L, Penny W. Dynamic causal modelling. NeuroImage. 2003;19(4):1273–1302

    2003

  2. [2]

    Granger causality analysis in neuroscience and neuroimag- ing

    Seth AK, Barrett AB, Barnett L. Granger causality analysis in neuroscience and neuroimag- ing. Journal of Neuroscience. 2015;35(8):3293–3297

    2015

  3. [3]

    A tutorial review of functional connectivity analysis methods and their interpretational pitfalls

    Bastos AM, Schoffelen JM. A tutorial review of functional connectivity analysis methods and their interpretational pitfalls. Frontiers in Systems Neuroscience. 2016;9:175

    2016

  4. [4]

    Open multimodal iEEG-fMRI dataset from naturalistic stimulation with a short au- diovisual film

    Berezutskaya J, Vansteensel MJ, Aarnoutse EJ, Freudenburg ZV , Piantoni G, Branco MP, et al. Open multimodal iEEG-fMRI dataset from naturalistic stimulation with a short au- diovisual film. Scientific Data. 2022;9:91

    2022

  5. [5]

    Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain

    Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage. 2002;15(1):273–289

    2002

  6. [6]

    A unified view of multitaper multivariate spectral estimation

    Walden AT. A unified view of multitaper multivariate spectral estimation. Biometrika. 2000;87(4):767–788

    2000

  7. [7]

    A convergence theorem for spectral factorization

    Wilson GT. A convergence theorem for spectral factorization. Journal of Multivariate Analysis. 1978;8(2):222–232

    1978

  8. [8]

    Sparse inverse covariance estimation with the graphical lasso

    Friedman J, Hastie T, Tibshirani R. Sparse inverse covariance estimation with the graphical lasso. Biostatistics. 2008;9(3):432–441

    2008

  9. [9]

    Calculation of the amount of information about a random function contained in another such function

    Gelfand IM, Iaglom A. Calculation of the amount of information about a random function contained in another such function. American Mathematical Society Translation Series. 1959;2:3–52

    1959

  10. [10]

    A nonparametric efficient evaluation of partial directed coherence

    Amblard PO. A nonparametric efficient evaluation of partial directed coherence. Biological Cybernetics. 2015;109(2):203–214

    2015

  11. [11]

    Sampling properties of the spectrum and coherency of sequences of action potentials

    Jarvis M, Mitra PP. Sampling properties of the spectrum and coherency of sequences of action potentials. Neural computation. 2001;13(4):717–749

    2001

  12. [12]

    Information theoretic interpretation of fre- quency domain connectivity measures

    Takahashi DY , Baccalá LA, Sameshima K. Information theoretic interpretation of fre- quency domain connectivity measures. Biological cybernetics. 2010;103:463–469. 20

    2010

  13. [13]

    Partial directed coherence: a new concept in neural structure determination

    Baccalá LA, Sameshima K. Partial directed coherence: a new concept in neural structure determination. Biological cybernetics. 2001;84(6):463–474

    2001

  14. [14]

    Nonparanormal graph quilting with applications to calcium imaging

    Chang A, Zheng L, Dasarathy G, Allen GI. Nonparanormal graph quilting with applications to calcium imaging. Stat. 2023;12(1):e623

    2023

  15. [15]

    Graph quilting: graphical model selection from partially observed covariances

    Vinci G, Dasarathy G, Allen GI. Graph quilting: graphical model selection from partially observed covariances. arXiv preprint arXiv:191205573. 2019

    2019

  16. [16]

    Low-rank matrix completion in the presence of high coherence

    Liu G, Li P. Low-rank matrix completion in the presence of high coherence. IEEE Trans- actions on Signal Processing. 2016;64(21):5623–5633

    2016

  17. [17]

    Deterministic symmetric positive semidefinite matrix completion

    Bishop WE, Yu BM. Deterministic symmetric positive semidefinite matrix completion. Advances in Neural Information Processing Systems. 2014;27

    2014

  18. [18]

    Estimation and Inference on Granger Causality in a Latent High- Dimensional Gaussian Vector Autoregressive Model

    Fan Y , Han F, Park H. Estimation and Inference on Granger Causality in a Latent High- Dimensional Gaussian Vector Autoregressive Model. Working Paper. 2020

    2020

  19. [19]

    Spontaneous behaviors drive multidimensional, brainwide activity

    Stringer C, Pachitariu M, Steinmetz N, Reddy CB, Carandini M, Harris KD. Spontaneous behaviors drive multidimensional, brainwide activity. Science. 2019;364(6437):eaav7893

    2019

  20. [20]

    Generalized partial directed coherence

    Baccala LA, Sameshima K, Takahashi DY . Generalized partial directed coherence. In: 2007 15th International conference on digital signal processing. Ieee; 2007. p. 163–166

    2007

  21. [21]

    Spectrum estimation and harmonic analysis

    Thomson DJ. Spectrum estimation and harmonic analysis. Proceedings of the IEEE. 2005;70(9):1055–1096

    2005

  22. [22]

    A large- scale standardized physiological survey reveals functional organization of the mouse visual cortex

    de Vries SE, Lecoq JA, Buice MA, Groblewski PA, Ocker GK, Oliver M, et al. A large- scale standardized physiological survey reveals functional organization of the mouse visual cortex. Nature neuroscience. 2020;23(1):138–151

    2020

  23. [23]

    Investigating causal relations by econometric models and cross-spectral meth- ods

    Granger CW. Investigating causal relations by econometric models and cross-spectral meth- ods. Econometrica: journal of the Econometric Society. 1969; p. 424–438

    1969

  24. [24]

    System identification

    Ljung L. System identification. In: Signal analysis and prediction. Springer; 1998. p. 163–173

    1998

  25. [25]

    Toward causal representation learning

    Schölkopf B, Locatello F, Bauer S, Ke NR, Kalchbrenner N, Goyal A, et al. Toward causal representation learning. Proceedings of the IEEE. 2021;109(5):612–634

    2021

  26. [26]

    Spatio- temporal correlations and visual signalling in a complete neuronal population

    Pillow JW, Shlens J, Paninski L, Sher A, Litke AM, Chichilnisky E, et al. Spatio- temporal correlations and visual signalling in a complete neuronal population. Nature. 2008;454(7207):995–999

    2008

  27. [27]

    Systematic errors in connectivity inferred from activity in strongly recur- rent networks

    Das A, Fiete IR. Systematic errors in connectivity inferred from activity in strongly recur- rent networks. Nature Neuroscience. 2020;23(10):1286–1296

    2020

  28. [28]

    shotgun

    Soudry D, Keshri S, Stinson P, Oh Mh, Iyengar G, Paninski L. Efficient" shotgun" inference of neural connectivity from highly sub-sampled activity data. PLoS computational biology. 2015;11(10):e1004464

    2015

  29. [29]

    Autoregressive modeling and causal ordering of economic variables

    Hsiao C. Autoregressive modeling and causal ordering of economic variables. Journal of Economic Dynamics and Control. 1982;4:243–259. 21

    1982

  30. [30]

    Causal inference in time series analysis

    Eichler M. Causal inference in time series analysis. Causality: Statistical perspectives and applications. 2012; p. 327–354

    2012

  31. [31]

    Inferring neural population dynamics from multiple partial recordings of the same neural circuit

    Turaga S, Buesing L, Packer AM, Dalgleish H, Pettit N, Hausser M, et al. Inferring neural population dynamics from multiple partial recordings of the same neural circuit. Advances in Neural Information Processing Systems. 2013;26

    2013

  32. [32]

    Integration of partially observed multimodal and multiscale neural sig- nals for estimating a neural circuit using dynamic causal modeling

    Kang J, Park HJ. Integration of partially observed multimodal and multiscale neural sig- nals for estimating a neural circuit using dynamic causal modeling. PLOS Computational Biology. 2024;20(12):e1012655

    2024

  33. [33]

    A causal network analysis of neuromodulation in the mood processing network

    Qiao S, Sedillo JI, Brown KA, Ferrentino B, Pesaran B. A causal network analysis of neuromodulation in the mood processing network. Neuron. 2020;107(5):972–985

    2020

  34. [34]

    Unsupervised identification of the internal states that shape natural behavior

    Calhoun AJ, Pillow JW, Murthy M. Unsupervised identification of the internal states that shape natural behavior. Nature neuroscience. 2019;22(12):2040–2049

    2019

  35. [35]

    Active neural coordination of motor behaviors with internal states

    Zhang YS, Takahashi DY , El Hady A, Liao DA, Ghazanfar AA. Active neural coordination of motor behaviors with internal states. Proceedings of the National Academy of Sciences. 2022;119(39):e2201194119

    2022

  36. [36]

    Analysis of dynamic brain imaging data

    Mitra PP, Pesaran B. Analysis of dynamic brain imaging data. Biophysical journal. 1999;76(2):691–708

    1999

  37. [37]

    A brain-wide map of neural activity during complex behaviour

    Angelaki D, Benson B, Benson J, Birman D, Bonacchi N, Bougrova K, et al. A brain-wide map of neural activity during complex behaviour. Nature. 2025;645(8079):177–191

    2025

  38. [38]

    Reproducibility of in vivo electrophysiological measurements in mice

    Laboratory IB, Banga K, Benson J, Bhagat J, Biderman D, Birman D, et al. Reproducibility of in vivo electrophysiological measurements in mice. Elife. 2025;13:RP100840

    2025

  39. [39]

    Functional connectomics spanning multiple areas of mouse visual cortex. Nature. 2025;640(8058):435–447

    2025

  40. [40]

    A mesoscale connectome of the mouse brain

    Oh SW, Harris JA, Ng L, Winslow B, Cain N, Mihalas S, et al. A mesoscale connectome of the mouse brain. Nature. 2014;508(7495):207–214

    2014

  41. [41]

    Spectral inference for high dimensional time series

    Zhang C, Zhang D. Spectral inference for high dimensional time series. IEEE Transactions on Information Theory. 2025;71(4):2909–2929

    2025

  42. [42]

    Restricted strong convexity and weighted matrix completion: Optimal bounds with noise

    Negahban S, Wainwright MJ. Restricted strong convexity and weighted matrix completion: Optimal bounds with noise. The Journal of Machine Learning Research. 2012;13:1665– 1697

    2012

  43. [43]

    Dimensionality reduction for large-scale neural recordings

    Cunningham JP, Yu BM. Dimensionality reduction for large-scale neural recordings. Na- ture neuroscience. 2014;17(11):1500–1509

    2014

  44. [44]

    Power spectra of general shot noises and Hawkes point processes with a random excitation

    Brémaud P, Massoulié L. Power spectra of general shot noises and Hawkes point processes with a random excitation. Advances in Applied Probability. 2002;34(1):205–222

    2002

  45. [45]

    Simple model of spiking neurons

    Izhikevich EM. Simple model of spiking neurons. IEEE Transactions on Neural Networks. 2003;14(6):1569–1572. 22

    2003

  46. [46]

    How directed is a directed network? Royal Society Open Science

    MacKay RS, Johnson S, Sansom B. How directed is a directed network? Royal Society Open Science. 2020;7(9):201138

    2020

  47. [47]

    Trophic coherence de- termines food-web stability

    Johnson S, Domínguez-García V , Donetti L, Muñoz MA. Trophic coherence de- termines food-web stability. Proceedings of the National Academy of Sciences. 2014;111(50):17923–17928

    2014

  48. [48]

    Theta oscillations in the hippocampus

    Buzsáki G. Theta oscillations in the hippocampus. Neuron. 2002;33(3):325–340

    2002

  49. [49]

    Mechanisms and functions of theta rhythms

    Colgin LL. Mechanisms and functions of theta rhythms. Annual Review of Neuroscience. 2013;36:295–312

    2013

  50. [50]

    Columnar organization in the midbrain periaqueductal gray: mod- ules for emotional expression? Trends in Neurosciences

    Bandler R, Shipley MT. Columnar organization in the midbrain periaqueductal gray: mod- ules for emotional expression? Trends in Neurosciences. 1994;17(9):379–389

    1994

  51. [51]

    Midbrain circuits for defensive behaviour

    Tovote P, Esposito MS, Botta P, Chaudun F, Fadok JP, Markovic M, et al. Midbrain circuits for defensive behaviour. Nature. 2016;534(7606):206–212

    2016

  52. [52]

    The neural control of vocalization in mammals: a review

    Jürgens U. The neural control of vocalization in mammals: a review. Journal of V oice. 2009;23(1):1–10

    2009

  53. [53]

    A specialized neural circuit gates social vocalizations in the mouse

    Tschida K, Michael V , Takatoh J, Han BX, Zhao S, Sakurai K, et al. A specialized neural circuit gates social vocalizations in the mouse. Neuron. 2019;103(3):459–472. Appendix A. Mathematical details of the SPIDER pipeline This appendix collects the equations underlying the narrative in Materials and Methods. Ap- pendix Appendix B and the following sectio...

    2019

  54. [54]

    non-hierarchicalness

    The process admits a V AR representation of orderp∈N∪ {+∞}if X(t)= pX s=1 A(s)X(t−s)+ε(t),(A.1) where theA(s) areK×Kmatrices and{ε(t)}is an innovation process withE[ε(t)ε(s) T ]=0 for t,sandE[ε(t)ε(t) T ]= Σ = Σ T . The autoregressive transfer function is ¯A(ω)=I− Pp s=1 A(s)e iωs forω∈[−π, π). The informational PDC from seriesℓto seriesk(Eq. (1) of the m...

    2020