{"id":"7d85afd5-9377-4775-bda1-e162c3339031","arxiv_id":"2510.12776","paper_version":4,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"Quantum-circuit sampling with hand-wired CNOT gates generated toy single-cell data whose programmed interactions Pearson/Spearman correlations missed; CellChat then flagged the very ligand-receptor pairs wired into the circuit.","lead":"qSimCells runs a tiny quantum circuit to generate synthetic single-cell gene-expression data, using CNOT gates as stand-ins for gene regulation and cell-cell signaling. The paper claims these data carry 'non-classical' dependencies that standard correlation tools cannot recover, and that CellChat confirms the programmed interactions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No evidence that the quantum kernel is 'essential' or that sampled dependencies are 'non-classical'; a classical generative baseline is never compared.","rationale":"The reader correctly identifies multiple flaws, but the single most load-bearing concern is the unsupported and empirically vacuous claim that the quantum kernel is essential. To justify the abstract's 'non-classical dependencies' and 'quantum kernel is essential', the paper must demonstrate that no classical generative model can produce the same distributional properties. It does not; it only shows that Pearson/Spearman are insufficient. This is a weak baseline. The chosen test directly compares a classical Bayes net with the same causal edges. Because the pipeline is a 10-qubit circuit executed on AerSimulator, the distribution is exactly a classical probability mass function; the quantum circuit is a specific parameterization. A classical Bayes net can parameterize the same distribution arbitrarily well. The CNOT-as-activation concern is real but narrower: even if the mapping were corrected, the essentiality claim would still fail. Hence I focus on the latter. The verdict remains REJECT because the overclaim is central and not fixable by minor editing; the paper would need to be re-scoped or provide a rigorous classical baseline. This aligns with the reader's final judgment, though the emphasis differs, so agreement is partial.","tokens_in":11764,"tokens_out":8210,"duration_ms":69043,"concrete_test":"Implement a classical generative baseline with the exact same directed graph: define binary variables g0..g9 with a Bayesian network whose edges are 3→5, 5→7, 7→0 (plus independent priors from Table 1). Use conditional probability tables chosen to reproduce the marginal activation probabilities of the quantum circuit (computed by statevector simulation). Generate the same number of cells, apply the same NB augmentation and HKGs, and repeat the Pearson/Spearman GRN inference and CellChat analysis with the same custom LR database. If the classical baseline exhibits the same failure to recover the programmed edges and similar relative increases for the 'true' LR pairs (e.g., ~75-fold for g7→g0), then the quantum kernel is not essential. If the classical model cannot reproduce the exact fold-change patterns or the correlation structures, that would provide evidence for a genuine quantum expr","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that qSimCells' quantum kernel produces 'non-classical dependencies' and is 'essential' for high-fidelity ground truth (Abstract; §4). This is the load-bearing assertion: if a classical generative model with the same directed structure reproduces the same phenomena, the quantum contribution is redundant. Three observations undermine it. (i) The experiments use AerSimulator, a classical simulation of a 10-qubit circuit; the output is a standard classical joint distribution over bitstrings. Entanglement in the statevector does not make computational-basis samples non-classical; no Bell/contextuality test is performed, and none can be from a single measurement setting. (ii) The only evidence for 'non-classical' behavior is that Pearson/Spearman correlations fail to recover the programmed edges (Fig. 4) and that CellChat gives large fold-changes for true LR pairs (Table 3). But any directed probabilistic model—e.g., a Bayesian network with edges 3→5, 5→7, 7→0—can produce correlations that are not recoverable by pairwise correlation, and will show LR-specific increases if the database contains those edges. (iii) The Discussion claims such causal dependencies are 'impossible to capture using classical, correlation-based simulators'—a straw man, since classical simulators like SERGIO/scMultiSim are not merely correlation-based. Consequently, the claim of quantum essentiality is an overreach with no comparative evidence.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces qSimCells, a quantum-computing-based simulator of single-cell transcriptomes. A parameterized quantum circuit with Ry rotations and CNOT gates is used to encode gene regulatory networks (intra-state) and cell-cell communication (inter-state) with explicit causal directionality. The circuit is sampled with a classical simulator (AerSimulator) to produce bitstrings, which are then transformed into negative binomial count matrices. The authors claim that the resulting synthetic data exhibit 'non-classical dependencies' that standard correlation-based tools cannot recover, that a CellChat analysis validates the programmed ligand-receptor pairs, and that the quantum kernel is 'essential' for high-fidelity benchmark data. The central claims are that the quantum approach uniquely enables joint modeling of gene-gene and cell-cell interactions and that it produces non-classical, causal ground-truth data.","tokens_in":12111,"tokens_out":5337,"duration_ms":45498,"significance":"If the claims held, qSimCells would be the first quantum-based simulator to jointly generate gene-regulatory and cell-cell communication ground truth with explicit causal directionality, with implications for benchmarking single-cell inference tools. The circuit formulation (Eqs. 1-4) is straightforward, the sampling pipeline is clearly described, and the code is publicly available. However, the central interpretive claims do not follow from the results. The output of a 10-qubit circuit measured in the computational basis is a classical joint distribution; the AerSimulator is itself a classical computation. The 'non-classical dependencies' claim is therefore unsupported in this setting. The CellChat 'validation' is circular because the 'true' LR pairs are exactly the CNOT edges programmed into the circuit. No classical generative baseline is compared, so the 'essential' role of the quantum kernel is not established. These are load-bearing issues that affect the abstract, results, and discussion.","major_comments":[{"comment":"The claim that the synthetic data exhibit 'non-classical dependencies' is unsupported and, in the present setup, incorrect. Equations (1)-(4) define a 10-qubit circuit; measuring in the computational basis (Section 2.3) yields a classical probability distribution over 1024 bitstrings. AerSimulator computes this distribution classically. Entanglement in the statevector does not make computational-basis samples non-classical, and no Bell/contextuality test or quantum-advantage argument is provided. The failure of Pearson/Spearman to recover the programmed CNOT edges (Fig. 4) is also a well-known property of any directed probabilistic model with confounding variables; it is not evidence of non-classicality. The abstract and Section 4 statements that the quantum kernel is 'essential' and that the data exhibit 'non-classical dependencies' therefore overreach.","section":"§2.1-2.3, Abstract, §3.5.1"},{"comment":"The CellChat validation is circular. The 'true' LR pairs (g3→g5 and g7→g0) are exactly the inter-state CNOT edges in L1 = {(3,5),(5,7),(7,0)} (Section 3.2.1, Eq. 4). Their relative increase in inferred communication probability when the inter-state interaction is 'active' is a direct consequence of the engineered circuit, not an independent confirmation of the model's biological validity. The 'false' pairs (g8→g4, g9→g4) were deliberately chosen as non-entangled, so their stability is also by construction. The results in Table 3 are therefore an internal consistency check, not a validation of the quantum kernel or of the proposed gene-activation semantics.","section":"§3.5.2, Table 3"},{"comment":"The mapping from CNOT to 'reinforced activation or deactivation' is unvalidated and not generally monotonic. For a Ry(θ) initialization, a CNOT(c,t) gate changes the target activation probability to p_t + p_c(1 - 2 p_t) (assuming independent qubits before the gate). This increases p_t only when p_t < 0.5 and decreases it when p_t > 0.5. The manuscript does not verify this condition for the programmed edges, yet it treats CX gates as 'explicit causal directionality' and as models of gene activation (Section 2.1, Section 3.2.1). The claimed 'causal ground truth' is thus inherited from an unvalidated analogy rather than established.","section":"§2.1, §3.2.1"},{"comment":"The Discussion asserts that the modeled cascades are 'impossible to capture using classical, correlation-based simulators' and implies that classical simulators are merely correlation-based. This is a straw man: SERGIO [5] and scMultiSim [6] are GRN-guided simulators and are not limited to pairwise correlation. No classical generative baseline (e.g., a Bayesian network with edges 3→5→7→0) is compared, so the claim that the quantum kernel is 'essential' lacks comparative evidence. In addition, the UMAP separation in Fig. 3 is confounded: both cases include structural sparsity (CT genes are zero in the other cell type, Section 3.4), so the difference between the two circuits cannot be attributed solely to the inter-state cascade.","section":"§4, §1, Fig. 3"}],"minor_comments":[{"comment":"The text says 'as described in Section 3.3' but the negative binomial augmentation is described in Section 2.3. Several other section-number cross-references should be checked.","section":"§3.3"},{"comment":"The fold-change for g7→g0 is stated as 'approximately 75-fold' in the Results and full abstract, while the top-level abstract (as provided) reports '~98-fold'. Table 3 gives 0.0830/0.0011 ≈ 75.5. Please harmonize.","section":"Abstract / §3.5.2"},{"comment":"All p-values are exactly 0. For a finite simulation this is not a meaningful significance measure; report uncertainty or use a non-zero threshold. The text itself later acknowledges that absolute p-values are misleading.","section":"Table 3"},{"comment":"The notation CX_{c,t} is used in the text but the figure does not clearly label control-target directions. Clarify the convention and the little-endian reversal with the global gene indices.","section":"Fig. 2 / §3.2"},{"comment":"Reference [27] concerns Bell-type non-classical correlations; citing it in support of 'non-classical dependencies' of computational-basis samples is not appropriate.","section":"§4"}],"recommendation":"reject","confidential_remarks":"The paper's central claim of quantum advantage / non-classicality is not supportable for a 10-qubit circuit sampled in the computational basis, and the validation is circular. The issues are load-bearing and cannot be fixed with minor edits. A reframing as a 'quantum-inspired' classical generative model with a proper classical baseline would be a different paper. Given the journal scope, rejection is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Alex,\n\nYou asked for my honest read on qSimCells. The practical core is a simple, working pipeline: prepare a 10-qubit product state, apply a handful of CNOTs along a programmed topology, sample with AerSimulator, and convert the bitstring histogram into expression counts via a negative binomial. That part is correct, and the repo appears to contain the code. As a toy ground-truth generator for 5-gene, 2-cell-type scenarios, it could be useful for stress-testing inference tools.\n\nThe problems start when the paper interprets the output. Calling the sampled correlations 'non-classical' is misleading: AerSimulator is a classical simulator, and the computational-basis distribution is an ordinary classical joint distribution over 10 bits. Entanglement in the statevector does not make the samples non-classical. There is no Bell-type test, and the paper offers no derivation. The abstract's 'no existing simulator' is false — scMultiSim (their own ref [6]) already models both GRNs and cell-cell interactions. The CellChat validation is circular: the 'true' LR pairs (g3→g5, g7→g0) are exactly the CNOT edges programmed in L1; their relative increase is a designed consequence, not an independent confirmation. Also note the abstract says 'up to 98-fold' while the text and Table 3 give 75-fold; that is a discrepancy the authors should catch.\n\nThe stress-test note is right: a classical directed graphical model with the same edges would produce correlations that Pearson/Spearman could miss and would give LR-specific increases if the database contains those edges. So the 'quantum kernel is essential' claim is not established. No classical baseline is compared.\n\nWould I cite this? No. Would I bring it to reading group? Maybe, to dissect how quantum-advantage claims can go wrong. My recommendation: the right outcome is a desk reject, or at most a 'major revision' request that forces the authors to re-scope the paper as a small benchmark generator and drop the non-classical/essentiality language. As written, the load-bearing interpretive claims fail, and the paper would waste referee time in its current state.","headline":"A functional toy quantum-circuit sampler, but the non-classical and quantum-essentiality claims are unsupported and the validation is circular.","tokens_in":12644,"tokens_out":2978,"would_cite":false,"duration_ms":25436,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A quantum circuit with CNOT gates can encode known gene-regulatory and cell-cell communication structure into synthetic single-cell data that classical correlation methods cannot recover.","keywords":["quantum generative modeling","single-cell RNA-seq simulation","gene regulatory networks","cell-cell communication","quantum entanglement","CNOT gates","synthetic transcriptomics","quantum kernel"],"falsifier":"Compute the exact marginal probabilities of the quantum circuit for each programmed edge (e.g., q3→q5) and check whether the target gene's activation probability moves in the direction the edge is supposed to impose when the control is ON versus OFF; if an edge programmed as an activation reduces the target's activation probability, the claimed causal ground truth does not correspond to the programmed regulation.","tokens_in":11571,"feed_emoji":"🧬","tokens_out":6587,"duration_ms":51730,"temperature":0.7,"pith_summary":"This paper introduces qSimCells, a quantum-computing-based simulator of single-cell RNA-seq data. Its central claim is that a parameterized quantum circuit—rotation gates for basal expression and a time-ordered sequence of CNOT gates for regulation—can encode both gene regulatory networks inside a cell type and ligand-receptor communication between cell types, so that the generated data carries explicit causal ground truth. If true, this would fill a gap the paper identifies: no existing simulator jointly models intra-cellular and inter-cellular interactions, and classical simulators rely on linear correlations. The paper argues that standard correlation analyses (Pearson and Spearman) do not recover the programmed causal paths, instead reporting spurious associations driven by high baseline expression probabilities, while a cell-cell communication inference tool detects the true ligand-receptor pairs through a large relative increase in communication probability.","feed_headline":"Quantum circuit builds synthetic cells with known gene wiring","feed_subtitle":"Classical correlation tools miss the encoded regulatory paths, giving benchmarks with known ground truth to test inference methods.","key_machinery":"The central object is the parameterized quantum circuit with R_y rotations and CNOT gates. The R_y rotations initialize basal gene activation; the CNOT gates entangle control and target gene qubits, with the sequence of control-target pairs explicitly programmed as the GRN and communication topology. The paper treats the CNOT as a directional activation/deactivation coupling—the mechanism that gives the simulator 'known causal ground truth'—and the tensor product of two cell-type registers creates the space in which inter-cell ligand-receptor entanglement is defined. The time-ordered application of CNOT gates is what produces a joint probability distribution that is not a simple sum of pairw","core_discovery":"On the paper's own terms, the discovery is a quantum kernel that maps gene-regulatory and cell-cell communication topologies onto a multi-qubit state. Each gene is a qubit whose R_y rotation encodes a basal activation probability; the ordered list of control-target pairs for CNOT gates defines which gene activates or deactivates which other gene, both within a cell type and across two cell-type registers combined by a tensor product. Sampling the final state produces binary gene-activation patterns, which are then turned into realistic counts by a negative-binomial transform. In a five-gene, two-cell-type proof of concept the paper programs a cascade q3→q5→q7→q0 and reports that classical co","pith_inferences":["Editorial inference: The paper's claim of 'non-classical dependencies' is about the distribution sampled from the circuit; a purely classical simulation of the same circuit would reproduce the same statistics, so any benchmark advantage lies in the circuit's expressive structure, not in hardware quantumness as such.","Editorial inference: The biological interpretation depends on whether a CNOT's XOR flip behaves like gene activation; testing the direction of the effect on target activation probability for each programmed edge would clarify whether the 'ground truth' is biologically meaningful.","Editorial inference: A natural extension is to compare qSimCells outputs against a classical generative model with the same pairwise activations; if classical correlation methods also fail on that, the failure is attributable to circuit-level nonlinearity rather than to entanglement specifically."],"forward_implications":["If qSimCells works as claimed, it provides benchmark datasets where the causal regulatory architecture is exactly known, letting developers test GRN-inference and cell-communication tools against ground truth rather than against guessed networks.","Classical correlation-based inference should be expected to fail on such data, producing spurious edges biased by highly expressed genes; users should not interpret correlation networks from quantum-generated ground truth as causal.","Relative changes in inferred communication probability between interacting and non-interacting conditions, not absolute probabilities or p-values, should be the readout when validating ligand-receptor pairs in simulated data.","Because the kernel is parameterized and supports execution on noisy quantum hardware, the same framework could generate larger, more realistic datasets as qubit counts and gate fidelities improve."],"fun_headline_variants":["Quantum simulator plants known gene circuits in synthetic cells","Quantum kernel creates cell data with encoded regulatory links","Synthetic cells with quantum-encoded gene interactions","Quantum model yields synthetic cells with known gene paths","Quantum circuit simulates cells with traceable gene regulation"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that a CNOT gate—a controlled flip of the target qubit whenever the control qubit is in state 1—is a faithful model of one gene activating or repressing another; if this mapping is wrong, the programmed 'causal ground truth' is not biological regulation.","fun_headline_variants_meta":{"raw":{"variants":["Quantum simulator plants known gene circuits in synthetic cells","Quantum kernel creates cell data with encoded regulatory links","Synthetic cells with quantum-encoded gene interactions","Quantum model yields synthetic cells with known gene paths","Quantum circuit simulates cells with traceable gene regulation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000399,"raw_usage":{"total_tokens":1937,"prompt_tokens":773,"completion_tokens":1164,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":517,"completion_tokens_details":{"reasoning_tokens":1106}},"tokens_in":517,"tokens_out":1164,"duration_ms":8875,"temperature":1.0,"reasoning_tokens":1106,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T09:52:22.076756+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the exact marginal probabilities of the quantum circuit for each programmed edge (e.g., q3→q5) and check whether the target gene's activation probability moves in the direction the edge is supposed to impose when the control is ON versus OFF; if an edge programmed as an activation reduces the target's activation probability, the claimed causal ground truth does not correspond to the programmed regulation.","supporting_citations":[],"review_version":1}