{"id":"07817be6-182a-4620-bd7b-8decbdbd8620","arxiv_id":"2412.17043","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"In a whole-brain mean field model, operating near a coupling-induced phase transition simultaneously maximizes signal propagation and diversity of intrinsic timescales.","lead":"Researchers built a second-order mean field model of the human brain from a spiking network with a real connectome and found a phase transition as cross-regional coupling increases. Near the critical point, visual signals propagate farther and brain regions show a wider range of response timescales, suggesting criticality as a unifying principle.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central criticality result is computed only in the reduced mean-field model; the spiking DTB is never tested for attenuation length or ACW diversity, and the known approximations (effective current, moment closure) shift the critical point, leaving the brain-level claim unvalidated.","rationale":"The paper has real independent support: the MNN reproduces mean firing rates across gamma, the phase transition is seen in both DTB and MNN, and the SI controls (linear connectome model, rewired network) make the joint roles of nonlinearity and structure more plausible. My concern is not that the model is wrong, but that the specific quantities used for the headline claim, attenuation length and ACW diversity, are never computed in the more biophysical DTB. The shift in gamma_c from 31.6 to 41.8 is explicitly attributed to approximations in the reduction; if those approximations also change the response near the transition, the unifying-mechanism claim is a statement about the MNN, not about the brain-like spiking network. The reader's CONDITIONAL verdict already captures this, so I recommend keeping the verdict unchanged until the DTB-level response check is reported.","tokens_in":24588,"tokens_out":9228,"duration_ms":94384,"concrete_test":"Run the same V1 pulse and white-noise protocols (Methods 4.6-4.7) in the DTB spiking model at gamma = 30.0, 31.6, 32.0, and 33.0, computing the attenuation length from the exponential fit E(d)=A exp(-d/Delta) and the ACW entropy/range ratio from the envelope method. If these quantities do not peak near gamma=31.6 or if the peak location disagrees with the spiking bifurcation, the mean field's 'optimal criticality' result does not transfer to the DTB. If full-scale DTB runs are infeasible, repeat at reduced neuron count with per-region in-degrees and connectome density preserved, and report finite-size scaling in the peak location.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing premise for the abstract's 'brain operating near criticality' claim is that the second-order moment-closure mean field (MNN) preserves the response statistics of the DTB spiking model near the transition, not just the phase boundary. The paper itself documents that the reduction is not exact: Results 2.2 places the DTB critical point at gamma_c=31.6 and the MNN critical point at gamma_c=41.8, and Methods 4.3 notes that the effective-current approximation inherits the boundary condition P(Vth)=0, which 'is not true when synaptic decay is present.' Yet all of the main signal-transmission and timescale-diversity evidence (Figs. 3D-3F and 4E-4G) is produced by simulating the MNN at gamma=36.0, 41.8, 42.8; the spiking DTB is only used to establish the phase transition and BOLD correlation. Nothing shows that the spiking model's attenuation length and ACW entropy peak near its own critical point. Because the mean-field reduction is a controlled approximation, the peaked metrics could be a property of the second-order closure rather than of the spiking dynamics it claims to coarse-grain. The secondary protocol asymmetry in Methods 4.7 (band-pass filtering only in the supercritical regime before computing ACW envelopes) makes the supercritical side of the ACW peak additionally sensitive to a hand-chosen filter.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a second-order moment-closure mean-field model (MNN) of the human brain by coarse-graining a 378-region spiking Digital Twin Brain (DTB) model with a DWI-derived connectome. The authors identify a phase transition driven by the cross-regional coupling strength γ, with the DTB critical point at γc=31.6 and the MNN critical point at γc=41.8. Using the MNN at subcritical (γ=36.0), critical (γ=41.8), and supercritical (γ=42.8) values, they report that a pulse stimulus delivered to V1 propagates with maximal attenuation length near criticality, and that Gaussian white-noise stimulation produces autocorrelation windows whose entropy and range ratio peak near criticality. They conclude that criticality jointly optimizes signal transmission and timescale diversity, and that the structural connectome scaffolds the resulting timescale hierarchy.","tokens_in":24779,"tokens_out":6594,"duration_ms":58290,"significance":"If the central claim holds, the paper provides a substantive bridge from a biophysically detailed spiking model to a tractable whole-brain mean-field model, and it offers a concrete mechanism—operating near a connectome-driven phase transition—for two often-separated phenomena: reliable signal propagation and hierarchical timescale diversity. The quantitative metrics (attenuation length, ACW entropy, range ratio) are measured from simulations rather than imposed, and the SI control experiments (a linear model on the same connectome and a nonlinear model on a rewired Erdős–Rényi network) correctly isolate the contributions of nonlinearity and structure. The analytic gradient of the moment activation and the use of automatic differentiation for the Jacobian are useful technical contributions. The main caveat is that the load-bearing evidence is generated entirely in the reduced mean-field model, whose fidelity to the spiking dynamics near the transition is asserted rather than demonstrated, so the significance for the real brain is conditional on that fidelity being established.","major_comments":[{"comment":"The central claim that the brain operates near criticality to achieve optimal signal transmission and timescale diversity is supported only by simulations of the MNN at γ=36.0, 41.8, and 42.8. The spiking DTB is used to locate the phase transition and to validate BOLD correlations, but its attenuation length and ACW diversity are never measured near its own critical point γc=31.6. Given that the MNN and DTB critical points differ by about 32%, and given the paper's own caveats (Methods 4.3: the effective-current approximation inherits P(Vth)=0, \"which is not true when synaptic decay is present\"; Methods 4.4: τ is fitted by matching oscillation frequency and each region is treated as homogeneous), the peaked metrics could in principle be an artifact of the second-order closure rather than a property of the spiking dynamics the model coarse-grains. Please either compute the attenuation length and ACW metrics in the DTB at γ≈31.6, or provide a direct validation that the MNN response statistics (not just the phase boundary) match the DTB in the critical region.","section":"§2.4–2.5, Figs. 3–4; Methods 4.3–4.4"},{"comment":"The ACW estimation is asymmetric across regimes: for the subcritical and critical regimes the envelope is computed from the raw response, whereas for the supercritical regime the response is first band-pass filtered with a rectangular window of width 10 Hz centered on an oscillation frequency estimated from one region (8Ad, right hemisphere). This makes the supercritical side of the ACW peak in Fig. 4(E) dependent on a hand-chosen filter. Since the claim is that entropy and range ratio \"increase sharply as the system approaches criticality\" and then drop in the supercritical regime, the comparison should be insensitive to this choice. Please report a sensitivity analysis over filter width and filter type, or apply the same filtering protocol to all regimes.","section":"§4.7, Fig. 4E"},{"comment":"The attenuation length Δ is extracted from an exponential fit E(d)=A e^{−d/Δ}, but the paper reports no goodness-of-fit or confidence intervals for Δ. In the critical regime the binned energy appears nearly flat over distance (Fig. 3F, left), in which case the exponential fit is poorly constrained and the peak in Δ near γc could reflect a fitting artifact. Please report fit residuals, confidence bounds, and a comparison with alternative decay forms (e.g., power law or constant) for the same data.","section":"§2.4, Fig. 3(F)"}],"minor_comments":[{"comment":"The phrase \"is found to induced\" should read \"is found to induce\".","section":"Abstract"},{"comment":"Calling the shift from γc=31.6 to γc=41.8 \"slightly shifted\" understates a roughly 32% change; please rephrase to be more precise.","section":"§2.2"},{"comment":"The stimulus is described as a \"20 mV current stimulus\"; current is not measured in mV. Please clarify the units and whether this is an injected current or a voltage offset.","section":"§4.6"},{"comment":"The notation h2β(Ēαi) is ambiguous; it should be written as (hβ(Ēαi))² to be consistent with Eq. (11).","section":"Eq. (10)"},{"comment":"The statement that in-degrees are the same for all neurons in a given region is in tension with SI S1.1, which describes cropping cross-regional in-degrees to a range [71, 3500]; please clarify how the coarse-grained K values are obtained.","section":"§4.4 and SI S1.1"},{"comment":"Panels (B)-(C) are single trials while (D)-(G) are averaged over 100 trials; please state whether the single-trial examples are representative and report trial-to-trial variability for the ACW metrics.","section":"Fig. 4"},{"comment":"There is a typo: \"possbile\" should be \"possible\".","section":"SI S3"},{"comment":"Reference 24 lists \"Xin Du\" and \"Jianfeng Feng\" twice in the author list; please correct the reference.","section":"References"},{"comment":"The GitHub link is appreciated; please also provide a versioned archive (e.g., Zenodo DOI) or a commit hash for reproducibility.","section":"Code availability"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of q-bio.NC and should be of interest to the criticality and whole-brain modeling community. The main risk is not circularity—the metrics are measured, not fitted—but the unvalidated transfer of mean-field criticality results to the spiking DTB. If the authors can add DTB-based measurements or a stronger equivalence argument, the paper could be suitable for publication. I would not reject on the basis of the current evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper deserves a serious referee. The genuinely new object is a second-order moment neural network for 378 human brain regions, derived from a conductance-based spiking digital twin brain model, and it shows a coupling-induced phase transition. The results are coherent: near the mean-field critical point at gamma=41.8, attenuation length peaks, ACW entropy and range ratio peak, and BOLD correlation with real fMRI is highest near criticality in both the DTB and the mean-field model. The SI control experiments (linear connectome model, rewired network) do real work in showing that nonlinearity and connectome jointly shape the timescale hierarchy.\n\nThe model derivation is substantial. The moment closure for conductance-based LIF neurons with synaptic decay uses an effective-current approximation, the Jacobian is computed with PyTorch autodiff plus analytic gradients, and code and data are public. The paper is also honest about the known caveat that the boundary condition P(Vth)=0 is not strictly true with synaptic decay, and it explicitly notes the critical-point shift between the DTB (gamma_c=31.6) and the MNN (gamma_c=41.8). I don't see a circularity problem: the attenuation length and ACW metrics are measured across gamma and peak near the model's critical point without being fitted to that peak.\n\nNow the soft spots. The central evidence for optimal signal transmission and timescale diversity comes only from simulating the mean-field model at gamma = 36.0, 41.8, and 42.8. The spiking DTB is used to establish the phase transition and to validate BOLD, but it is never used to measure attenuation length or ACW diversity. Because the reduction shifts the critical point by about 30%, it remains an open question whether the spiking model's response statistics actually peak near its own critical point. The abstract says \"brain operating near criticality\"; the evidence supports \"mean-field model operating near criticality.\" Also, tau is fitted to match the spiking oscillation frequency, each region is treated as homogeneous, and the supercritical ACW pipeline applies a hand-chosen band-pass filter only in that regime. The supercritical side of the ACW peak is therefore partly a function of that filtering choice. There are no error bars in the key figures.\n\nMy judgment: the central modeling result is likely right and is a genuine contribution. The \"unifying mechanism for the brain\" claim needs either DTB-based validation of at least attenuation length near its critical point, or a clear statement that this is a property of the reduced model. That is a revision-level issue, not a desk-reject.\n\nThis paper is for computational neuroscientists interested in whole-brain models, the critical brain hypothesis, and timescale hierarchy. I would bring it to reading group. Recommendation: send it to peer review, and ask the authors to address the DTB validation gap, filter sensitivity, and uncertainty quantification.","headline":"A serious whole-brain mean-field study with a real criticality story, but the headline brain claim outruns the evidence because the spiking model is never tested on the main metrics.","tokens_in":25415,"tokens_out":2431,"would_cite":true,"duration_ms":24249,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that near a connectome-driven phase transition, a whole-brain mean field model simultaneously minimizes signal attenuation and maximizes diversity of intrinsic timescales, proposing criticality as a unifying mechanism for…","keywords":["criticality","mean field model","intrinsic timescales","signal transmission","whole brain model","autocorrelation window","phase transition","structural connectome"],"falsifier":"Simulate the full Digital Twin Brain spiking model at its own critical point gamma = 31.6 and measure the fitted attenuation length of a V1 pulse response and the entropy/range ratio of ACWs across regions; if these quantities do not peak near gamma = 31.6, or peak only in the mean field reduction at gamma = 41.8, the claim that criticality simultaneously optimizes both properties in the brain model fails.","tokens_in":24294,"feed_emoji":"🧠","tokens_out":4687,"duration_ms":41114,"temperature":0.7,"pith_summary":"This paper tries to establish that criticality—the state a whole-brain model reaches when cross-regional coupling crosses a phase transition—simultaneously solves two problems the brain faces: transmitting sensory signals across the cortex without fading, and producing a wide range of intrinsic timescales across regions. The authors build a second-order mean field model from a large-scale spiking 'digital twin brain' model on the human structural connectome, and show that as the global coupling strength approaches the critical point, the attenuation length of a visual stimulus diverges while the entropy and range ratio of autocorrelation windows peak. If true, this gives a single dynamical mechanism by which the brain can be both sensitive to input and temporally diverse.","feed_headline":"Critical state maximizes signal reach and timescale diversity","feed_subtitle":"A connectome-grounded brain model finds one coupling threshold that both minimizes signal decay and broadens timescales.","key_machinery":"The central object is a second-order mean field model, called a moment neural network, obtained by moment closure (Fokker-Planck diffusion approximation) and coarse-graining of a conductance-based spiking Digital Twin Brain model with 378 HCPex regions and DWI-derived structural connectivity. The state is the per-region mean and variance of excitatory and inhibitory firing rates, evolved by tau dm/dt = -m + Phi(m), with Phi decomposed into synaptic summation, an effective-current approximation for conductance-based synapses, and a moment activation mapping. Linear stability analysis of the Jacobian reveals the phase transition and its eigenmodes; signal transmission is quantified by response energy and attenuation length, and timescale diversity by the entropy and range ratio of autocorrelation-window (ACW) estimates of envelope responses.","core_discovery":"The paper's central claim is that near the critical point of a cross-regional coupling-induced phase transition, the model attains optimal signal transmission and maximal timescale diversity simultaneously. Concretely, a brief step stimulus to V1 loses less response energy across the cortex near gamma_c = 41.8 (mean field) than below or above it, with fitted attenuation length maximal at the critical point; under white-noise stimulation, the distribution of autocorrelation windows across regions broadens from narrow and unimodal to broad and multimodal, and both entropy and range ratio diverge near criticality. The authors argue criticality is a unifying mechanism: structural connectome provides the scaffold, nonlinearity allows the phase transition, and operating near the transition amplifies the hierarchy of timescales while minimizing attenuation.","pith_inferences":["A direct empirical prediction is that in vivo measures of evoked-response attenuation and ACW diversity should covary across individuals or states: whichever condition moves the brain closer to criticality should show both longer attenuation lengths and broader timescale distributions.","The same mechanism may generalize beyond vision: any focal input, such as auditory or somatosensory stimulation, should show distance-dependent attenuation that is minimal near criticality, a testable extension of the V1-centred analysis.","The phase-shift response used for the supercritical regime suggests a second signature of criticality: near the transition, perturbation response should become slow and scale-invariant, which could be probed with time-resolved perturbation experiments.","Because the critical point differs between the spiking model (gamma = 31.6) and the mean field approximations (gamma = 41.8), quantitative comparisons to empirical data should rely on the spiking critical point; the mean field point is a qualitative proxy."],"forward_implications":["Near criticality, a focal visual stimulus propagates across the whole brain with minimal attenuation, so higher-order regions receive usable signal despite being far from the input site.","The same critical state produces a broad, multimodal distribution of intrinsic timescales, with short timescales in early visual areas and long timescales in transmodal regions such as area 46, OFC, and ACC.","The structural connectome supplies the scaffold for the timescale hierarchy, while biophysical nonlinearity lets criticality amplify it without requiring extreme coupling strengths.","The mean field model predicts the DTB spiking model's phase transition at a shifted critical point and reproduces BOLD correlations peaking near the critical regime, so the mechanism is testable at the macroscale.","Beyond the critical point, signal energy attenuates sharply and timescales become homogeneous, implying the brain must regulate cross-regional coupling near the transition to keep both properties."],"supporting_citations":[{"why":"Supplies the Digital Twin Brain spiking model, the parcellation, and the data-assimilated external inputs from which the mean field model is derived.","marker":"[23, 24]"},{"why":"Provides the effective-current approximation that lets conductance-based LIF neurons with synaptic decay be handled by current-based moment activation.","marker":"[44]"},{"why":"Gives the efficient numerical algorithm for moment activation and its gradients, making whole-brain stability analysis tractable.","marker":"[43]"},{"why":"Establishes the structural-connectome mechanism for hierarchical timescales that this paper extends with criticality.","marker":"[13]"},{"why":"Shows hierarchical connectome modes and critical states maximizing functional diversity, the comparison point for the eigenmode analysis.","marker":"[27]"},{"why":"Provides the Fokker-Planck diffusion approximation and first-passage-time moment closure underlying the moment activation mapping.","marker":"[42, 40]"},{"why":"Defines the autocorrelation window (ACW) used to measure intrinsic timescales from response envelopes.","marker":"[29]"}],"fun_headline_variants":["Criticality boosts signal reach and timescale diversity","One critical threshold optimizes brain signal and timing","Brain near criticality: best signal, richest timescales","Optimal signal and timescale spread at brain criticality"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the second-order moment closure with the effective-current approximation captures the spiking dynamics faithfully enough near the phase transition, so that the divergent attenuation length and timescale diversity found in the mean field model are properties of the spiking brain model and not artifacts of the reduction.","fun_headline_variants_meta":{"raw":{"variants":["Criticality boosts signal reach and timescale diversity","One critical threshold optimizes brain signal and timing","Brain near criticality: best signal, richest timescales","Optimal signal and timescale spread at brain criticality"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000169,"raw_usage":{"total_tokens":1235,"prompt_tokens":883,"completion_tokens":352,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":288}},"tokens_in":499,"tokens_out":352,"duration_ms":3934,"temperature":1.0,"reasoning_tokens":288,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T05:50:42.497408+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the full Digital Twin Brain spiking model at its own critical point gamma = 31.6 and measure the fitted attenuation length of a V1 pulse response and the entropy/range ratio of ACWs across regions; if these quantities do not peak near gamma = 31.6, or peak only in the mean field reduction at gamma = 41.8, the claim that criticality simultaneously optimizes both properties in the brain model fails.","supporting_citations":[{"cited_title":"Becker and Marco A","cited_arxiv_id":null,"evidence_quote":"Provides the effective-current approximation that lets conductance-based LIF neurons with synaptic decay be handled by current-based moment activation."},{"cited_title":"Moment neural network and an efficient numerical method for modeling irregular spiking activity","cited_arxiv_id":null,"evidence_quote":"Gives the efficient numerical algorithm for moment activation and its gradients, making whole-brain stability analysis tractable."},{"cited_title":"Hierarchical timescales in the neocortex: Mathematical mechanism and biological insights","cited_arxiv_id":null,"evidence_quote":"Establishes the structural-connectome mechanism for hierarchical timescales that this paper extends with criticality."},{"cited_title":"Hierarchical connectome modes and critical state jointly maximize human brain functional diversity","cited_arxiv_id":null,"evidence_quote":"Shows hierarchical connectome modes and critical states maximizing functional diversity, the comparison point for the eigenmode analysis."},{"cited_title":"A hierarchy of intrinsic timescales across primate cortex","cited_arxiv_id":null,"evidence_quote":"Defines the autocorrelation window (ACW) used to measure intrinsic timescales from response envelopes."}],"review_version":1}