{"id":"46ddbdf8-db33-450c-b5b8-e2ddd694c19f","arxiv_id":"2506.06514","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Quantum random walks outperform classical random walks in disease gene prioritization and cell-cell communication analysis, but the reported advantage rests on selectively chosen parameters and a fixed default baseline.","lead":"This paper tests whether quantum random walks rank disease genes and map cell communication pathways better than classical random walks. In benchmarks across five gene networks and three diseases, the quantum walk appears to lead on the headline metric, but the comparison is uneven because the quantum time parameter is optimized while the classical baseline is left at its default.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unfair comparison: CTQRW's time parameter is tuned on the test ground truth while RWR uses a fixed default restart probability, so the claimed ranking advantage is not established.","rationale":"The paper's central quantitative claim is that CTQRW outperforms RWR in ranking disease genes. The reader identified the fairest and most load-bearing weakness: CTQRW's continuous time parameter is optimized directly on the ground-truth target genes, while RWR is left at a fixed default. This is a textbook example of test-set tuning, and it alone invalidates the comparison. The paper provides no variance estimates or statistical tests, so the reported differences cannot be separated from noise. The cell-cell interaction result is anecdotal and does not rescue the main claim; it is a qualitative observation without a quantitative benchmark. I agree with the reader's verdict that the current evidence does not support 'more accurately rank disease-associated genes.' A properly re-run comparison with validation-based model selection for both methods would settle whether the advantage is real, which is why my concrete test focuses on that protocol. No other concern in the manuscript outweighs this one; the rest of the analysis is descriptive and would be worthwhile only after the primary comparison is made fair.","tokens_in":9169,"tokens_out":3237,"duration_ms":34854,"concrete_test":"Use a validation split: for each disease-network pair, partition target genes into tuning and test sets. Select t for CTQRW and alpha for RWR by maximizing AP@K on the tuning set only, then evaluate both on the held-out test set. Repeat with 100 random splits and report bootstrap 95% CIs for the difference; if the CI covers zero for most networks, the superiority claim is an artifact of test-set tuning.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Central claim of Section III-A/Fig. 2 is undermined by selection bias. CTQRW's time t is swept and the maximum mean AP@K is reported, while RWR uses NetworkX's default PageRank alpha (0.85) with no tuning. Since t is selected on the same target genes used as ground truth, CTQRW is effectively tuned on the test set; RWR is not. No error bars, replicate seeds, or significance tests accompany the differences (e.g., Autism 0.3 vs 0.14). Thus the claimed 'significantly outperformed' is not established; a fair comparison could eliminate the advantage.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates the use of quantum random walks (QRWs) in two biomolecular network analysis tasks: continuous-time quantum random walks (CTQRW) for disease gene prioritization on five interactome networks across asthma, autism, and schizophrenia; and discrete-time quantum random walks (DTQRW) for analyzing a cell-cell interaction (CCI) network from mouse brown adipose tissue. The central claims are that CTQRW ranks disease-associated genes more accurately than classical random walk with restart (RWR), and that DTQRW identifies biological pathways (e.g., CD8+ T-cell to malignant-cell signaling) that classical discrete-time random walks overlook. The paper reports mean AP@20 improvements, maximum AP@K comparisons, and a qualitative community analysis of the CCI network.","tokens_in":9280,"tokens_out":5673,"duration_ms":59544,"significance":"If the claims were properly established, this would be a meaningful contribution to the emerging literature on quantum walk applications in network medicine, extending earlier work on quantum disease-gene prioritization (e.g., Saarinen et al.) to a broader set of interactomes and a novel multipartite CCI setting. The use of publicly available GWAS-derived networks and a real scRNA-seq-based CCI network is a strength, and the paper explicitly acknowledges some limitations (e.g., dead ends in multipartite walks). However, the current evaluation methodology is not sound enough to support the headline claims: the central quantitative comparison is unfair, the CCI analysis is anecdotal, and no reproducibility artifacts (code, parameter grids, or statistical tests) are provided. The potential significance is therefore moderate, contingent on a substantially revised evaluation.","major_comments":[{"comment":"The central performance comparison is unfair and the claim that CTQRW 'significantly outperformed' RWR is not established. CTQRW's evolution time t is swept and the maximum mean AP@K over t is reported, while RWR uses the NetworkX default restart probability (presumably the standard PageRank α = 0.85) without any tuning. Moreover, the time t is selected on the same target genes that are used as ground truth for evaluation, so the comparison includes a form of test-set tuning. No error bars, replicate seeds, or significance tests accompany the reported differences (e.g., the Autism AP@20 values of 0.30 vs. 0.14 could be within noise). A fair comparison would tune RWR's α over a comparable range, choose both hyperparameters using a validation set or internal cross-validation, and report means and variances across repeats.","section":"Section III-A and Figs. 1–2"},{"comment":"The 'maximum metric AP@K' is an oracle metric that inflates apparent performance. Reporting the best AP@K at any time t does not describe the behavior of a fixed, usable algorithm, and it is not a standard evaluation protocol. The paper should either report performance as a function of t (as in Fig. 1) with a principled selection criterion, or average over a reasonable range of t; if a maximum is retained, it must be compared against a classically tuned baseline selected under identical conditions.","section":"Section III-A"},{"comment":"The CCI analysis is qualitative and anecdotal. The claim that DTQRW identifies 'key driver genes' overlooked by classical walks is based on a single example (the CD8+ T to malignant-cell path via L-Glutamine and SLC3A2) and a visual inspection of heatmaps. No quantitative metric, null model, or statistical test is applied to the transition-probability matrices or the resulting communities, so the claim that DTQRW is more sensitive to network structure is not supported. The analysis should define an evaluation criterion (e.g., enrichment of validated ligands/receptors, or a comparison of recovered communities against known biology) and evaluate both methods under identical conditions.","section":"Section III-B"},{"comment":"The implementation of CTQRW is underspecified, which prevents reproducibility and makes the comparison opaque. The paper does not state whether the Hamiltonian is the adjacency matrix A or the Laplacian L, whether the periodic wave-function collapse described in Section II-E is actually implemented, or whether the chiral Hamiltonian of Eq. (6) is used. The range and resolution of the time parameter t in the sweep are also not given. These details are essential for interpreting the results and for allowing others to repeat the experiments.","section":"Section II-E and Section III-A"}],"minor_comments":[{"comment":"The last paragraph of the introduction contains a sentence fragment: 'Recently, with the advent of quantum algorithms in pre-fault tolerant quantum hardware and their applications in biomedicine [3], clinical trials [4], and single-cell analyses [5], among others.' This should be completed or merged with the following sentence.","section":"Section I"},{"comment":"In Eq. (8), the summation index k conflicts with the free target index in p_{j->k}(t); the expression should be rewritten, for example as p_{j->k}(t) = Σ_{k'=1}^{d_j} |ψ_{j,k'}(t)|^2.","section":"Eq. (8)"},{"comment":"The caption contains a typographical error: 'IS' in 'The l2 distance ... IS represented' should be lowercase 'is'.","section":"Fig. 4 caption"},{"comment":"The name 'Bioplex3' should be spelled consistently as 'BioPlex3' throughout the table and text.","section":"Table I"},{"comment":"The phrase 'significantly better' is used colloquially without any statistical significance test; please replace it with a quantitative statement or add appropriate statistical tests.","section":"Section III-A"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central claim of quantum-walk superiority rests on an unfair comparison (tuning CTQRW's time parameter on the test set while leaving RWR's restart parameter at a fixed default) and on the absence of any uncertainty quantification. These are load-bearing issues that can be fixed with additional experiments and a revised evaluation protocol, so I do not recommend rejection at this stage; however, the current evidence does not support the advertised conclusions. The CCI section would also need a quantitative component before it can support the 'key driver genes' claim. I would also encourage the authors to release code and the exact parameter grids to improve reproducibility."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the paper is an honest, clearly written extension of known methods, but the headline claim that QRWs rank disease genes better than classical random walks is not supported by the experiments as designed. The CTQRW time parameter is swept and the best value is read off the same ground truth used for evaluation, while RWR gets the NetworkX default restart probability. That is not a fair head-to-head.\n\nWhat is actually new: a benchmark of continuous-time quantum walks (already proposed by Saarinen et al.) across five networks and three diseases, and a first application of discrete-time quantum walks to a cell-cell interaction network. The math is standard and correctly presented, and the authors are careful to cite Saarinen et al. and the prior network propagation literature. I see no citation problems; self-citation is not an issue here because the relevant prior work is correctly credited.\n\nThe main soft spot is the comparison. Figures 1 and 2 report the maximum mean AP@K over the CTQRW time parameter. For RWR they fix the restart probability at 0.85, untuned. No error bars, no replicates, no statistical tests. The statement \"significantly outperformed\" is not backed by any significance testing. A fair comparison would tune RWR's restart probability in the same way, or use cross-validation for both. With that change, the advantage may well disappear.\n\nThe CCI section is qualitative and anecdotal: one validated pathway and a heatmap. It is suggestive but not conclusive. The paper acknowledges some limitations, such as dead ends in multipartite graphs, but does not address the central evaluation problem.\n\nWho is this for? Researchers working on network-based disease gene prioritization who want to see a broad empirical test of quantum walk methods. The paper could be a useful benchmark if the methodology were fixed.\n\nI would not accept the paper as is. But it deserves a serious referee: the extension is useful, the data are real, and the flaws are methodological and fixable. I would send it to review with the expectation of major revision.","headline":"A useful extension undermined by an unfair baseline comparison.","tokens_in":9789,"tokens_out":1854,"would_cite":false,"duration_ms":21107,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A quantum random walk ranks disease genes at the top more often than a classical walk, and reveals a cell path classical diffusion misses.","keywords":["quantum random walks","network propagation","disease gene prioritization","biomolecular networks","cell-cell interaction networks","continuous-time quantum walk","discrete-time quantum walk","brown adipose tissue"],"falsifier":"Re-run the five-network comparison with RWR's restart probability optimized over a grid and with CTQRW's time parameter either averaged or selected by cross-validation; if the top-20 AP@K advantage over RWR shrinks below the reported margins or reverses, the central claim fails. A permutation test that shuffles the seed/target assignment and recomputes the maximum AP@K difference would show whether the reported gap is larger than chance, since the paper provides no error bars or significance test.","tokens_in":8989,"feed_emoji":"🧬","tokens_out":9113,"duration_ms":86064,"temperature":0.7,"pith_summary":"This paper tries to establish that quantum random walks—the interference-based, unitary analogues of classical diffusion on a graph—are a workable tool for analyzing real biomolecular networks, not just a theoretical construct. It reports that a continuous-time quantum walk ranks known disease-associated genes higher than a classical random walk with restart in the top-20 lists for asthma, autism, and schizophrenia across most of five gene networks. It also reports that a discrete-time quantum walk on a mouse brown-adipose cell-cell interaction network recovers a path from CD8+ T cells to malignant cells through L-glutamine and the SLC3A2 transporter, a route the classical walk does not find. If these results hold, quantum walks would give network medicine a concrete way to sharpen gene prioritization and to generate cell-communication hypotheses.","feed_headline":"Quantum walks beat classical walks on disease genes","feed_subtitle":"On five gene networks and a brown-adipose cell map, quantum walkers improve top-20 rankings and find a CD8+ T cell path.","key_machinery":"The load-bearing object is the quantum random walk on a graph whose Hamiltonian is the network itself. In continuous time, the walker evolves by the Schrödinger equation with $H=A$ or $H=L$, the initial amplitudes are normalized gene scores, and a gene's rank score is the squared transition amplitude $|\\langle j|e^{-iHt}|k\\rangle|^2$; because amplitudes add before being squared, paths interfere, which is the mechanism that makes the quantum ranking differ from classical diffusion. In discrete time, the walker lives on directed edge states $|j\\to k\\rangle$ and moves by a coin-and-shift unitary $U=SC$, making the dynamics deterministic and reversible on a superposition of paths. The continuous time parameter $t$ is the free knob the paper turns to select the maximum AP@K used in the comparison.","core_discovery":"The paper's central claim is that replacing classical diffusion with a unitary quantum walk changes which nodes of a biological network are highlighted, and that the change is biologically informative. For disease-gene prioritization, the paper compares a continuous-time quantum random walk (CTQRW), defined by $|\\psi(t)\\rangle=e^{-iHt}|\\psi(0)\\rangle$ with transition probability $p_{j\\to k}(t)=|\\langle j|e^{-iHt}|k\\rangle|^2$ and $H$ the adjacency or Laplacian matrix, against a classical random walk with restart (RWR) on the giant components of five gene networks. The paper reports that, allowing the evolution time $t$ to vary, CTQRW reaches higher mean AP@K at $K=20$ on most networks for all three diseases (for example about 0.7 versus a lower RWR value for asthma on HumanNet, and about 0.3 versus 0.14 for autism on PCNet), while the gap narrows at $K=100$. For the cell-cell interaction setting, the paper reports that a discrete-time quantum random walk (DTQRW) on a four-partite graph derived from mouse brown adipose tissue does not settle into the same community structure as the classical discrete-time walk, and that it finds a short path from CD8+ T cells to malignant cells via L-glutamine and SLC3A2 that the classical walk does not.","pith_inferences":["A natural stress test is to average CTQRW's AP@K over time instead of taking its maximum; if the averaged curve still beats RWR, the ranking advantage is robust, whereas if it does not, the practical gain hinges on choosing $t$ well in advance.","The reported CD8+ T-cell to malignant-cell path through L-glutamine and SLC3A2 is a concrete experiment: perturb SLC3A2 expression or glutamine availability in an adipose-tumor co-culture and ask whether the signaling route inferred from the network changes as the quantum walk predicts.","The time dependence of CTQRW's ranking quality could itself be a signal: peaks in AP@K over $t$ may mark characteristic propagation scales of a disease module, giving a new way to compare module geometry across diseases."],"forward_implications":["At $K=20$, CTQRW gives higher mean AP@K than RWR on most of the five networks across asthma, autism, and schizophrenia, and the advantage shrinks as $K$ grows to 100, so the practical gain is concentrated in the highest-ranked genes.","Because the comparison runs on the giant component of each network, the quantum walker's advantage is not an artifact of a single network size or density; it appears on both densely connected networks like PCNet and fragment-prone networks like BioPlex3 and STRING.","In the CCI network, DTQRW identifies L-glutamine, SLC3A2, and GABA as key metabolites in pathways the classical walk misses, which means quantum walks can be used to propose specific, literature-checkable cell-communication hypotheses."],"supporting_citations":[{"why":"It introduces the continuous-time quantum walk scoring idea for disease-gene prioritization that this paper applies.","marker":"[8]"},{"why":"It supplies the GWAS seed and target gene curation, the three diseases, and the five gene networks that define the benchmarking task.","marker":"[11]"},{"why":"It provides the single-cell RNA-seq data and the metabolite-mediated CCI network construction used for the mouse brown adipose tissue analysis.","marker":"[17]"},{"why":"It supplies the Hamiltonian-based transition-probability formalism for continuous-time quantum walks used to compute the rank scores.","marker":"[19]"},{"why":"It provides the classical random-walk-with-restart implementation used as the baseline in the disease-gene comparison.","marker":"[25]"},{"why":"It is the published evidence that IFN-gamma from CD8+ T cells downregulates SLC3A2, supporting the DTQRW-found path.","marker":"[27]"},{"why":"It is the published evidence for glutamine metabolism effects on CD8+ T cells used to interpret the found path.","marker":"[28]"},{"why":"It is the published evidence tying GABA signaling in brown adipose tissue to obesity, matching a DTQRW-identified metabolite.","marker":"[29]"}],"fun_headline_variants":["Quantum walks edge out classical in disease gene ranking","Quantum random walks sharpen biomolecular network analysis","Quantum walkers spot disease genes classical methods miss","Quantum walk finds CD8+ T cell route in fat tissue","Quantum walks improve network medicine predictions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that comparing the quantum walk's best performance over its time parameter with the classical walk's default restart setting is a fair head-to-head; if the classical walk were tuned as generously, the reported advantage could disappear.","fun_headline_variants_meta":{"raw":{"variants":["Quantum walks edge out classical in disease gene ranking","Quantum random walks sharpen biomolecular network analysis","Quantum walkers spot disease genes classical methods miss","Quantum walk finds CD8+ T cell route in fat tissue","Quantum walks improve network medicine predictions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000274,"raw_usage":{"total_tokens":1697,"prompt_tokens":1059,"completion_tokens":638,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":675,"completion_tokens_details":{"reasoning_tokens":568}},"tokens_in":675,"tokens_out":638,"duration_ms":6401,"temperature":1.0,"reasoning_tokens":568,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:54:18.446427+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the five-network comparison with RWR's restart probability optimized over a grid and with CTQRW's time parameter either averaged or selected by cross-validation; if the top-20 AP@K advantage over RWR shrinks below the reported margins or reverses, the central claim fails. A permutation test that shuffles the seed/target assignment and recomputes the maximum AP@K difference would show whether the reported gap is larger than chance, since the paper provides no error bars or significance test.","supporting_citations":[{"cited_title":"Disease gene prioritization with quantum walks","cited_arxiv_id":null,"evidence_quote":"It introduces the continuous-time quantum walk scoring idea for disease-gene prioritization that this paper applies."},{"cited_title":"Network propagation for gwas analysis: a practical guide to leveraging molecular networks for disease gene discovery","cited_arxiv_id":null,"evidence_quote":"It supplies the GWAS seed and target gene curation, the three diseases, and the five gene networks that define the benchmarking task."},{"cited_title":"Mebocost: Metabolite-mediated cell communication modeling by single cell transcriptome","cited_arxiv_id":null,"evidence_quote":"It provides the single-cell RNA-seq data and the metabolite-mediated CCI network construction used for the mouse brown adipose tissue analysis."},{"cited_title":"Enhanced quantum transport in chiral quantum walks","cited_arxiv_id":null,"evidence_quote":"It supplies the Hamiltonian-based transition-probability formalism for continuous-time quantum walks used to compute the rank scores."},{"cited_title":"Cd8+ t cells regulate tumour ferroptosis during cancer immunotherapy","cited_arxiv_id":null,"evidence_quote":"It is the published evidence that IFN-gamma from CD8+ T cells downregulates SLC3A2, supporting the DTQRW-found path."},{"cited_title":"Differential effects of glutamine inhibition strategies on antitumor cd8 t cells","cited_arxiv_id":null,"evidence_quote":"It is the published evidence for glutamine metabolism effects on CD8+ T cells used to interpret the found path."},{"cited_title":"Gamma-aminobutyric acid signaling in brown adipose tissue promotes systemic metabolic derangement in obesity","cited_arxiv_id":null,"evidence_quote":"It is the published evidence tying GABA signaling in brown adipose tissue to obesity, matching a DTQRW-identified metabolite."}],"review_version":1}