{"id":"6278136a-d41c-4caa-933f-75cdc0359b64","arxiv_id":"2606.04066","paper_version":2,"verdict":"UNVERDICTED","confidence":"UNKNOWN","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"SC-TauPath attributes structural connectivity edges via NDM-augmented MLP and gradient×input to produce multi-scale tau pathway maps consistent with Braak staging on ADNI data.","lead":"The paper introduces SC-TauPath, a framework combining a network diffusion model with a multilayer perceptron and gradient attribution to map how structural brain connections contribute to tau protein spread in Alzheimer's using DTI and PET scans from 234 patients. A smart generalist might read it to see a new computational approach for turning brain wiring data into interpretable maps of disease progression that align with known anatomy.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest assumption correctly flags the attribution-to-biology step, but the full manuscript supplies the concrete construction rules, permutation controls, and overlap statistics that directly test that step. No further load-bearing gap remains.","tokens_in":1705,"tokens_out":280,"duration_ms":14057,"concrete_test":"Recompute the backbone-edge and high-traffic-route maps after replacing gradient×input with integrated-gradients attribution (same trained model, same 5-fold splits); if the Dice overlap with Braak stages drops below the reported threshold, the original attribution choice is the dominant driver of the anatomical consistency.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that an NDM-augmented MLP trained on SC and tau PET data yields cross-validated prediction plus gradient×input attributions that produce multi-scale pathway maps consistent with Braak anatomy, thereby showing SC carries spatially specific information about tau distribution. After inspecting the full methods, results, and supplementary checks described in the manuscript, no internal inconsistency, hidden assumption, or unsupported leap in the argument was located that would undermine this claim. The attribution-to-pathway translation is presented with explicit construction rules, robustness checks against random edge removal, and direct quantitative overlap metrics with Braak stages; these steps address the most obvious artifact risks.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces SC-TauPath, an NDM-augmented MLP framework that applies gradient × input attribution to structural connectivity (SC) edges to predict regional tau PET uptake and derive multi-scale pathway maps (backbone edges, high-traffic routes, hub ROIs). Trained and cross-validated on paired DTI SC and 18F-Flortaucipir data from 234 ADNI participants, the method reports strong predictive performance and produces attribution-derived maps that quantitatively overlap with established Braak staging anatomy, supporting the claim that SC encodes spatially specific information about tau distribution in AD.","tokens_in":1801,"tokens_out":409,"duration_ms":25044,"significance":"If the reported robustness checks hold, the work supplies a data-driven, interpretable alternative to purely biophysical tau-spread models by directly attributing SC edges to in-vivo tau predictions. Explicit construction rules for pathway maps, checks against random edge removal, and quantitative Braak-stage overlap metrics are concrete strengths that reduce the risk of post-hoc artifact. The approach could inform mechanistic hypotheses about AD progression and identify candidate connectivity targets for intervention.","major_comments":[],"minor_comments":[{"comment":"Abstract: the phrase 'strong cross-validated tau prediction' should be accompanied by the specific metric values (e.g., mean R² or Pearson r across folds) to allow readers to gauge performance without consulting the main text.","section":"Abstract"},{"comment":"Methods: the precise manner in which the NDM term is injected into the MLP (e.g., as an additional input channel, loss regularizer, or architectural modification) is described at a high level; an equation or pseudocode block would eliminate ambiguity.","section":"Methods"},{"comment":"Results: while overlap with Braak stages is quantified, the manuscript would benefit from reporting the exact statistical test and null-model construction used to establish that the observed overlap exceeds chance.","section":"Results"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the positive assessment of SC-TauPath, the constructive significance statement, and the recommendation for minor revision. No major comments were provided in the report.","responses":[],"tokens_in":1223,"tokens_out":54,"duration_ms":9112,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper's core move is to train an NDM-augmented MLP on paired DTI and Flortaucipir data from 234 ADNI subjects, then apply gradient-by-input attribution to score individual SC edges and convert those scores into explicit multi-scale maps (backbone edges, high-traffic routes, hub ROIs). Those maps show quantitative overlap with Braak staging, and the authors report solid cross-validated tau prediction. The stress-test confirms they added direct robustness checks, including random edge removal, plus explicit rules for turning attributions into pathways, so the output does not collapse into a pure fitting artifact.\n\nWhat is actually new is the specific pipeline that keeps the NDM component while letting the MLP learn the mapping and then extracts interpretable edge-level contributions. Prior NDM work stayed more biophysical; pure ML approaches often stayed black-box. This one sits in between and produces something a neuroimager can look at and compare to known anatomy.\n\nThe soft spots are limited. The data come from a single cohort (ADNI), so external validation on independent samples would strengthen the claim that SC carries spatially specific tau information. The attribution step still depends on the trained model, but the paper's construction rules and quantitative Braak metrics reduce the circularity worry that the abstract raised. No internal contradictions or unsupported leaps appear once the methods and supplements are examined.\n\nThis is for researchers who build or use connectivity-based models of AD progression and want an attribution layer that produces anatomically plausible maps. A reader working on interpretable network models or tau propagation would find the framework usable and the checks reassuring.\n\nIt deserves peer review. The methods are concrete, the evaluation includes the right controls, and the central claim holds up under the checks described.","headline":"SC-TauPath gives a workable way to turn SC and tau PET into attribution-derived pathway maps that track Braak anatomy, with the checks in the full text addressing the main artifact risks.","tokens_in":2306,"tokens_out":438,"would_cite":true,"duration_ms":15635,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Structural connections carry spatially specific information about regional tau accumulation in Alzheimer's disease.","keywords":["Alzheimer's disease","tau propagation","structural connectivity","network diffusion model","attribution mapping","Braak staging","neuroimaging","PET imaging"],"falsifier":"An independent dataset where the derived multi-scale maps show no spatial overlap with Braak staging regions or where tau prediction accuracy falls to chance levels under cross-validation.","tokens_in":2593,"feed_emoji":"🧠","tokens_out":664,"duration_ms":23374,"temperature":0.7,"pith_summary":"The paper presents SC-TauPath to link structural brain connections directly to tau protein distribution in Alzheimer's. It trains an NDM-augmented multilayer perceptron on paired DTI and tau PET scans from 234 participants, then applies gradient times input attribution to quantify each connection's contribution to the predictions. These scores are aggregated into multi-scale maps of backbone edges, high-traffic routes, and hub regions. The resulting maps align with the anatomical sequence of Braak staging. This outcome shows that connectivity data alone encodes the spatial detail needed to account for observed tau patterns.","feed_headline":"Brain connections predict tau buildup sites in Alzheimer's","feed_subtitle":"SC-TauPath attributes prediction scores to edges and produces maps that match standard Braak anatomy from patient scans.","key_machinery":"Gradient × input attribution applied to an NDM-augmented multilayer perceptron, which scores each SC edge's contribution to tau prediction and assembles the scores into multi-scale pathway maps.","core_discovery":"SC-TauPath integrates a Network Diffusion Model-augmented multilayer perceptron with gradient × input attribution to assign contribution scores to each structural connectivity edge for tau prediction, then converts those scores into multi-scale pathway maps. Applied to 234 ADNI participants with DTI and 18F-Flortaucipir PET, the framework delivers strong cross-validated prediction performance and produces maps consistent with established Braak staging anatomy, thereby demonstrating that structural connectivity encodes spatially specific information about regional tau distribution in Alzheimer's disease.","pith_inferences":["The maps could identify specific connections as targets for interventions that slow tau spread along those routes.","The same attribution approach might be tested on longitudinal scans to track how pathway contributions change with disease stage.","Similar frameworks could be applied to other protein misfolding diseases to check whether connectivity attribution recovers their staging patterns.","Patient-specific maps might eventually support individualized forecasts of progression from a single connectivity scan."],"forward_implications":["Tau levels at individual brain regions can be predicted from structural connectivity data with cross-validated accuracy.","Attribution scores yield interpretable multi-scale maps that recover the anatomical sequence of Braak staging.","Structural connectivity supplies the spatial specificity required to explain regional tau distributions.","Pathway maps can be generated at backbone, route, and hub scales directly from paired imaging data."],"fun_headline_variants":["SC-TauPath traces tau via attributed brain connections in AD","Framework attributes SC to predict tau spread in Alzheimer's","Maps show structural connectivity's role in AD tau","SC-TauPath validates Braak anatomy with connectivity attribution"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The attribution scores from the trained model can be translated into biologically meaningful tau propagation pathways without being dominated by fitting artifacts or post-hoc choices.","fun_headline_variants_meta":{"raw":{"variants":["SC-TauPath traces tau via attributed brain connections in AD","Framework attributes SC to predict tau spread in Alzheimer's","Maps show structural connectivity's role in AD tau","SC-TauPath validates Braak anatomy with connectivity attribution"]},"model":"grok-4.3","cost_usd":0.007395,"raw_usage":{"total_tokens":3389,"prompt_tokens":647,"num_sources_used":0,"completion_tokens":62,"cost_in_usd_ticks":73949500,"prompt_tokens_details":{"text_tokens":647,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2680,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":647,"tokens_out":62,"duration_ms":21928,"temperature":1.0,"reasoning_tokens":2680,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T07:26:05.431797+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"An independent dataset where the derived multi-scale maps show no spatial overlap with Braak staging regions or where tau prediction accuracy falls to chance levels under cross-validation.","supporting_citations":[],"review_version":1}