PREreview of Zigzag Persistence of Neural Responses to Time-Varying Stimuli

summary: 'Zigzag Persistence of Neural Responses to Time-Varying Stimuli by Yuri Gardinazzi Alessio Ansuini Eugenio Piasini Fabio Anselmi and Matteo Biagetti proposes a pipeline that transforms frame-wise neuronal activity from the Sensorium 2023 mouse visual cortex dataset into evolving topological objects and analyzes them with zigzag persistent homology. Using persistence landscapes of H1 features as compact descriptors the study shows high clustering accuracy for repeated presentations of individual videos modest discrimination of video families and near-chance mouse identification. Temporal order and spatial contiguity are demonstrated as critical via shuffling controls. The work positions topological data analysis as a tool for uncovering time-resolved neural coding structure.'

keywords: 'zigzag persistence, persistent homology, persistence landscapes, topological data analysis, neural manifolds, mouse visual cortex, Sensorium 2023, two-photon calcium imaging, cubical complexes, simplicial complexes, intersection zigzag, clustering, adjusted rand index, logistic regression, video stimuli, time-varying stimuli, calcium imaging, neural population activity'

score: 75

tier: 'Tier2 (Graduate journals): Solid methodological novelty and clear controls, but limited robustness analyses, modest statistical reporting, and incomplete comparisons to strong non-topological baselines likely preclude Tier3+; nonetheless suitable for a graduate-level venue.'

CPI: 0.65

expected_citations_2yr: 13

categories:

Abstract:

score: 8

description: 'Self-contained and clear: objectives dataset methods (zigzag persistence with landscapes) and principal results (A/B/C) are stated; could benefit from reporting representative effect sizes and sample scope across all mice directly in the abstract.'

Recency:

score: 9

description: 'References include 2022–2026 works (Sensorium 2024 ICML 2025 recent neuroscience applications) alongside foundational TDA; recency is strong.'

Scope:

score: 9

description: 'Content aligns with the title and implied keywords: time-varying neural responses zigzag persistence and interpretable topological signatures; scope is focused and appropriate.'

Relevance:

score: 8

description: 'Addresses a pertinent gap (dynamic topology of neural activity) with interpretable descriptors; novelty is credible though comparison to simpler temporal features could be expanded.'

'Factual Errors':

score: 9

description: 'No substantive factual errors detected; methodological steps and metrics (e.g. ARI PCA logistic regression) are described consistently with standard practice.'

Language:

score: 8

description: 'Technical writing is clear and precise; minor stylistic/spacing artifacts in citations and symbols do not impede comprehension.'

Formatting:

score: 7

description: 'Overall consistent; minor LaTeX spacing footnote placement and equation rendering quirks reduce polish; cross-referencing to appendices is appropriate.'

Suggestions:

score: 7

description: 'New idea is present (intersection–zigzag adapter with landscapes for H1). Consider: (i) direct cubical zigzag vs. simplicial adapter comparisons (ii) multi-threshold superlevel sets (iii) inclusion of H0/H2 and images-to-topology baselines (iv) kernel or Riemannian alternatives for landscape vectorization (v) benchmarking against dynamic manifold methods.'

Problems:

score: 8

description: 'Addresses the limitation of static persistence for dynamic neural data; demonstrates sensitivity to temporal order and spatial contiguity. Practical significance for classification is modest which is noted and contextualized.'

Assumptions:

score: 7

description: 'Key assumptions include δ>0 thresholding 2D-plane interpolation and 3-simplex closure with 2-skeleton retention; these are stated and partially justified though their effect on cycle formation/closure warrants systematic ablation.'

Consistency:

score: 8

description: 'Findings are internally consistent across mice/conditions and align with expected controls (performance collapses under shuffling).'

Robustness:

score: 6

description: 'Good sanity checks (temporal/spatial shuffles) but limited sensitivity analysis over thresholds landscape layers sampling resolution PCA dimension and intersection vs. union zigzag variants; no cross-dataset validation.'

Logic:

score: 8

description: 'Conclusions follow from presented data: strong per-video clustering moderate family-level signal weak mouse identity; causal claims are avoided.'

'Statistical Analysis':

score: 7

description: 'Reports ARI mean±sd over resamples and CV accuracies; could be strengthened with confidence intervals permutation tests against chance multiple-comparisons control and effect-size reporting for classifier performance.'

Controls:

score: 8

description: 'Temporal-order and spatial-contiguity controls are well designed and interpretable; additional controls (e.g. non-topological temporal features) would sharpen attribution.'

Corrections:

score: 6

description: 'Normalization to δ is clear but potential confounders (arousal/behavior neuropil contamination plane-to-plane gain) are not modeled; mixed-effects or nuisance regression could reduce unexplained variance.'

Range:

score: 7

description: 'Covers multiple mice and several video families but parameter-space exploration (thresholds landscape sampling H-dimensions) is narrow; exploring a wider stimulus/parameter range would increase generality.'

Collinearity:

score: 6

description: 'Landscape features across layers/time may be correlated; PCA helps but no explicit multicollinearity diagnostics (e.g. VIF) are provided.'

'Dimensional Analysis':

score: 'N/A'

description: 'Physical dimensional analysis is not applicable to the mathematical constructs and set-based operations used here.'

'Experimental Design':

score: 7

description: 'Pipeline is reproducibly outlined with algorithmic steps and software. Recommend preregistered splits seed control explicit hyperparameters/regularization and stronger baselines (e.g. spectral/temporal CNNs optical-flow H0/H2 cubical persistence) to contextualize gains.'

'Ethical Standards':

score: 'informational'

description: 'Data reuse involves mouse two-photon imaging (Sensorium 2023). Please add a statement confirming compliance with the dataset’s original IACUC/ethical approvals and clarify that only de-identified secondary data were used with links to dataset license/ethics documentation.'

'Conflict Of Interest':

score: 'informational'

description: 'No conflicts are declared. Include a formal COI statement for all authors (e.g. funding affiliations software/IP related to fastzigzag/Dionysus2 or competing commercial interests).'

Normalization:

score: 'informational'

description: 'The study uses per-plane per-video mean-centering to define δ and threshold at zero. Consider alternative normalizations: per-neuron z-scoring across all videos robust scaling (median/IQR) per-plane variance normalization and temporal detrending/deconvolution sensitivity analyses.'

'Improve Citability':

score: 'informational'

description: 'To maximize reuse: (i) release code versioned environments and exact preprocessing scripts; (ii) publish precomputed per-plane and concatenated descriptors for all mice with standardized train/validation/test splits; (iii) provide ablation suite (thresholds landscape layers sampling H0/H1/H2 intersection vs. union zigzag cubical vs. simplicial adapter); (iv) add formal definitions and a concise theorem/complexity analysis for the adapter’s effect on H1; (v) include a baseline zoo (static persistence dynamic graph metrics optical-flow+PCA temporal CNN) with a model card; (vi) document random seeds CV protocols and confidence intervals; (vii) supply an API for plug-in descriptors to new datasets.'

'Idea Incubator':

score: 'informational'

description: '- Economics (business cycles as topological loops): Cyclical expansions/contractions map to birth/death of H1 features over time; policy shocks resemble frame shuffles disrupting temporal coherence collapsing identifiable cycles.

'- Systems control (observability/controllability): Persistence of loops corresponds to modes that remain observable under time-varying inputs; intersection zigzag approximates time-local observability windows.

'- Population biology (predator–prey cycles): Lotka–Volterra oscillations manifest as recurrent co-activation loops; parameter changes (stimulus families) shift cycle prominence akin to changing carrying capacities.

'- Statistical physics (percolation and phase transitions): Superlevel-set thresholding resembles percolation; the emergence/filling of cycles tracks connectivity phases; shuffles perturb correlation length and reduce persistent structures.

'- Information theory (redundancy–synergy trade-offs): H1 features approximate synergistic multi-neuron coordination; temporal shuffling reduces synergy measurable via diminished landscape amplitudes.

'- Network queuing (traffic wave formation): Congestion waves create loop-like flow patterns; stimulus-driven load variations analogize to time-dependent arrival rates shaping persistence landscapes.'

Falsifiability:

score: 'informational'

description: 'Primary claims: (1) Zigzag H1 landscapes distinguish repeated presentations of distinct videos for a given mouse; (2) Temporal order and spatial contiguity are critical; (3) Descriptors modestly encode video family and weakly encode mouse identity. Falsifiers: (a) Frame-shuffled or grid-scrambled data achieving similar ARI/accuracy as intact data; (b) Strong non-topological temporal baselines (e.g. optical-flow PCA or temporal CNN on δ-grids) matching/exceeding performance without topological features; (c) Replication on independent datasets showing near-chance per-video clustering under intact data; (d) Parameter-insensitive performance (landscape layers/thresholds) suggesting artifacts rather than topology carry the signal.'

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they used generative AI to come up with new ideas for their review.