{"id":"3f7f7dbf-a563-4835-ad58-b0ea6c6a0913","arxiv_id":"2501.13142","paper_version":1,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":0.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A perspective review concludes that computational biomedicine has matured and that mechanistic and data-driven modeling should be integrated rather than opposed.","lead":"This paper reviews how computer models of the heart and other biological systems have changed since 2000, and argues that mechanistic models and data-driven machine learning should be combined. It is useful as a field-level orientation for researchers, clinicians, and funders working on digital twins, simulated clinical trials, and AI in medicine.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claim that M&S has matured into an integral part of physiological/medical research is supported mainly by publication counts and selected success stories; the authors concede no data quantify M&S's contribution to health outcomes, so the maturity narrative is under-evidenced.","rationale":"The reader's weakest_assumption correctly identifies that the maturation narrative depends on publication growth and selected success stories, while the paper itself concedes it cannot quantify M&S's contribution to improved health outcomes. My stress-test confirms this is the most load-bearing concern: the central claim is empirical and testable, but the evidence cited is mostly activity-based and anecdotal. I agree with the reader's assessment. I do not find a separate technical or internal-consistency flaw; the paper is balanced about challenges (e.g., small-data paradox, model validation, FAIR barriers) and appropriately frames the synergy of mechanistic and data-driven approaches as a potential rather than an established fact. Because this is a review/perspective rather than a research preprint, and the reader's UNVERDICTED verdict already reflects that status, my concern does not move the verdict. The proposed systematic review would provide a quantitative check on whether the maturity claim overstates current integration of M&S into clinical and regulatory practice.","tokens_in":23271,"tokens_out":2819,"duration_ms":32538,"concrete_test":"Conduct a systematic review of clinical adoption: identify all prospective clinical trials and regulatory decisions (e.g., FDA or EMA) between 2000 and 2024 in which an in silico model served as primary evidence for safety or efficacy, using ClinicalTrials.gov, FDA approval documents, and EMA public assessment reports. If the count is small (say, fewer than 20) or dominated by the paper's own cited examples (CiPA, the two digital-twin ablation trials), then the assertion that M&S has become an 'integral part' of physiological and medical research is not supported by the available evidence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central assertion (abstract, Section 6) is that computational modelling has 'matured' and 'has become an integral part of physiological and medical research.' The evidence offered in Section 2 consists of (i) a five-fold increase in the PubMed ratio of M&S-related cardiology papers to 2.2% in 2023, (ii) selected success stories (Physiome/VPH, CiPA, digital twins, MedalCare-XL), and (iii) rising EU life expectancy. The authors explicitly concede that 'we are not aware of data that would allow one to quantify the contribution of M&S to this improvement' when linking M&S to the life-expectancy increase. This is a load-bearing gap: publication counts measure research activity, not clinical adoption or patient benefit, and the selected success stories are not demonstrated to be representative—several (MedalCare-XL, openCARP, the atrial shape model) come from the authors' own groups. Without evidence that M&S is routinely used in clinical workflows, regulatory decisions, or quantified health improvements, the conclusion that the field has matured into an integral part of biomedical research rests on anecdote and activity metrics. The synergy claim (mechanistic + data-driven) is more defensible because it is framed as a forward-looking potential and the paper discusses known limitations such as the 'plausibility trap' and domain gaps. No internal inconsistency was found; the concern is an evidentiary one about the strength of the central narrative.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This perspective article revisits the 2000 vision of Kohl, Noble, Winslow, and Hunter on computational modelling of biological systems, with a focus on the cardiovascular system. It argues that computational modelling and simulation (M&S) has matured into an integral part of physiological and medical research, that the old reductionism/integrationism debate has been superseded by a mechanistic-vs-data-driven dichotomy, and that the two approaches can and should be combined synergistically. The paper surveys recent developments in digital twins, in silico clinical trials, machine learning, open tools and standards, and selected cardiology applications, and it concludes with near-term frontiers and community-oriented recommendations. The article is explicitly a perspective/review and does not present new quantitative analyses.","tokens_in":23539,"tokens_out":4165,"duration_ms":47801,"significance":"If the central narrative is accepted, this paper provides a useful synthesis and agenda for the field, connecting historical visions to current tooling and identifying actionable needs such as FAIR data, metadata and software standards, uncertainty quantification, and better incentives for sharing. Its strengths include an extensive and largely accurate reference base, a clear description of the Physiome/VPH lineage, and an unusually honest treatment of limitations: the 'plausibility trap' in synthetic training data, domain gaps between simulated and real-world data, missing evidence for population-level health impact, and unresolved issues in digital twin personalisation. The paper's main value is as a statement of current consensus and future priorities by three leaders in the field, rather than as an empirical demonstration of clinical impact. Its central claims therefore need to be phrased with a level of confidence commensurate with the evidence presented.","major_comments":[{"comment":"The sentence 'The millennial promise of reducing morbidity and mortality turned out to be true: life expectancy in the EU rose...' is not supported by the evidence that follows, because the authors immediately concede that 'we are not aware of data that would allow one to quantify the contribution of M&S to this improvement.' This assertion is load-bearing for the paper's broader maturity-and-benefit narrative. I recommend either replacing this passage with concrete adoption evidence (e.g., the ICH E14/S7B update, regulatory qualification examples, or prospective clinical trials using model-derived predictions) or explicitly softening the claim to state that anticipated benefits are becoming plausible and are beginning to materialize in specific contexts, rather than that the promise has been fulfilled.","section":"Section 2 (life-expectancy paragraph)"},{"comment":"The assertion that M&S 'has become an integral part of physiological and medical research' is supported mainly by a five-fold increase in a PubMed fraction (reaching 2.2% in 2023) and by selected success stories, several of which originate from the authors' own groups (e.g., MedalCare-XL, openCARP, the bi-atrial statistical shape model). Publication counts measure research activity, not clinical integration or patient benefit, and the examples are not demonstrated to be representative of the field as a whole. Please add indicators of adoption beyond publication metrics (regulatory decisions, clinical guidelines, reimbursement, or numbers of patients reached) or reframe the conclusion as 'increasingly established in research and emerging in specific clinical contexts,' which would be better matched to the evidence actually presented.","section":"Section 2 opening and Section 6"},{"comment":"The statement that digital twins 'have proven valuable across diverse medical applications' is stronger than the cited reviews support. The two prospective trials referenced later in Section 4 ([140,141]) are encouraging but small and restricted to ablation guidance, and the other cited sources are position papers or reviews. Because clinical utility is central to the maturity narrative, I recommend replacing 'have proven valuable' with 'have shown promise in early clinical evaluations' or 'are being evaluated in prospective clinical settings.'","section":"Section 2.2 (Digital Twin Approaches)"}],"minor_comments":[{"comment":"The phrase 'in silico generated data' should be hyphenated as 'in silico-generated data' for consistency with standard usage.","section":"Section 2.4"},{"comment":"There are two consecutive 'Acknowledgments' blocks after the author contributions; these should be merged into a single section.","section":"Section 3"},{"comment":"The sentence stating that in silico research has an 'absence of inherent variability, obviating the need for repeated experiments' is overstated for numerical simulations, which still face discretization error and stochastic algorithm variability; the later discussion of uncertainty quantification should be reflected in this sentence as well.","section":"Section 2.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a perspective by leading figures in the field and contains numerous self-citations (e.g., [20,39,70,72,78,114]); for a review this is acceptable, though the reliance on the authors' own examples for some central claims reinforces the need for softening or for independent supporting evidence. The duplicate acknowledgments section suggests the manuscript would benefit from a final editorial pass. The paper is better suited to a review or perspective venue than to an original-research journal, since it presents no new quantitative results; the main question for the editor is whether the confidence levels in the abstract and conclusion can be aligned with the evidence base before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a perspective, not a research paper—no new equations, data, or algorithms—and the reader is right to treat it as unverdictable on novelty grounds. What it does well is synthesize twenty-five years of computational cardiac modeling and map the old 'mountain and village' debate onto today's mechanistic-vs-data-driven tension. That framing is not original (it draws on Corral-Acero and others), but it is useful and clearly explained. The paper is also honest in places you don't usually see in perspective pieces: it flags the 'plausibility trap,' domain gaps in synthetic data, and the absence of any data linking M&S to the rise in life expectancy.\n\nThe soft spots are real but not disqualifying. The central assertion that M&S has 'matured' and become 'an integral part' of physiological and medical research is asserted rather than demonstrated. The evidence is a five-fold increase in a PubMed fraction plus selected success stories, several from the authors' own groups (openCARP, MedalCare-XL, Cobiveco). That is fine for a perspective, but it is advocacy, not proof, and a skeptical reader should discount the selection. The life-expectancy passage is overreach: even with the hedge, it invites a causal reading that the authors themselves disavow. If I were editing, I would cut that sentence or reframe it. Self-citation is heavy but mostly legitimate—they are leading actors in the field—still, the bibliography reads at times like a highlight reel of their own software and datasets. Minor editorial sloppiness: duplicate acknowledgments section and duplicated references (e.g., [82]=[72], [98]=[80], [60]=[56]) suggest a hasty final pass.\n\nThe stress-test concern about the maturity claim is fair, but the paper's transparency about the missing evidence keeps it from being a load-bearing flaw. The synergy claim is forward-looking and argued with appropriate caution. No internal contradictions.\n\nWho benefits: newcomers to cardiac M&S who want a map of the field, and established researchers wanting a quick update on digital twins, in silico trials, and ML/M&S integration. It is not a research contribution. I would send it to peer review as a perspective, with a request to temper the 'integral part' wording and clean up references. The core message is defensible and useful.","headline":"A useful, well-referenced perspective on 25 years of cardiac M&S that is honest about its limits; the 'maturity' claim is asserted more than proven, but this is advocacy, not fraud.","tokens_in":24055,"tokens_out":3565,"would_cite":false,"duration_ms":35541,"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":"The paper argues that computational modelling of biological systems has matured into an integral part of biomedical research, and that the next advances will come from deliberately combining mechanistic, first-principles models with…","keywords":["computational modelling","in silico medicine","digital twins","in silico clinical trials","mechanistic modelling","machine learning","cardiac electrophysiology","FAIR data"],"falsifier":"A head-to-head comparison in which machine-learning classifiers trained on synthetic data from mechanistic cardiac models are outperformed, on out-of-sample clinical data, by classifiers trained on equal-sized clinical datasets for the same diagnostic task would undermine the synergy claim. So would a post-market audit showing that in silico trial predictions—for instance, proarrhythmia risk classifications—fail to correlate with adverse outcomes in subsequent real clinical trials.","tokens_in":23053,"feed_emoji":"🫀","tokens_out":8115,"duration_ms":83289,"temperature":0.7,"pith_summary":"This review article argues that computational modelling and simulation (M&S) of biological systems has matured over the past quarter century into an integral part of basic and clinical research, with cardiology as the leading example. The authors claim that the old reductionism-versus-integrationism debate has largely settled, and that the newer split between mechanistic, first-principles models and data-driven machine learning is a false dichotomy: the two approaches complement each other, because mechanistic models generalise and respect physical laws while data-driven models stay anchored to real observations. They document this maturity in concrete forms: personalised digital twins, synthetic virtual cohorts (\"digital chimeras\") that make in silico clinical trials possible, synthetic data that trains machine-learning classifiers, and open standards that make models interchangeable. The stake is practical: if the review is right, further progress in computational medicine depends less on choosing between physics-based and data-driven methods than on deliberately integrating them, backed by shared data and quality-controlled workflows.","feed_headline":"Mechanistic models and AI belong together, cardiology review argues","feed_subtitle":"Simulation has matured into a clinical tool; the next leap pairs first-principles models with machine learning.","key_machinery":"The conceptual centrepiece is the \"mountain and village\" pairing, integrationist panorama versus reductionist close-up, which the authors repurpose into the modern pairing of mechanistic (village) and data-driven (mountain-top) modelling. The concrete machinery that carries the technical claims is the multi-scale cardiac electrophysiology hierarchy: ion-channel kinetics described by ordinary differential equations, cell-level models that couple currents into action potentials, tissue-level reaction-diffusion partial differential equations that propagate excitation, and Poisson's equation mapping the resulting fields to a body-surface electrocardiogram. This hierarchy is what makes digital twins personalisable, digital chimeras sampleable, and synthetic labelled training data producible at scale. Around it, the paper places an ecosystem layer of markup languages, FAIR data and software practices, and credibility frameworks, which is the mechanism that makes models interoperable and trustworthy in regulatory settings.","core_discovery":"On its own terms, the paper's central claim is that \"in silico methods\" have become \"an integral part of physiological and medical research\" and that the next wave of progress will come from a synergy between mechanistic and data-driven modelling. The review traces a shift away from the early ambition of one universal model of \"the human\" toward two complementary targets: digital twins, which personalise a baseline mechanistic model to a single patient, and digital chimeras, synthetic individuals sampled from shape and parameter distributions so that statistically representative cohorts can undergo simulated interventions. It further claims that machine learning does not replace mechanistic modelling but is strengthened by it: validated multi-scale simulations supply large, class-balanced, precisely labelled synthetic training data; physics-informed neural networks constrain learning with known laws; and hybrid simulated-plus-clinical datasets improve classifiers in atrial fibrillation, flutter, fibrosis, and electrolyte disorders. The paper's corollary is that the binding constraints on future progress are data standardisation, uncertainty quantification, reproducibility, and community infrastructure rather than computational power.","pith_inferences":["A fair quantitative test of the maturation narrative would be a longitudinal analysis of how many M&S publications in cardiology report prospective clinical implementation, as opposed to retrospective simulation studies; the ratio would show whether adoption is broad or confined to showcases.","If the synergy claim generalises, the same digital-chimera-plus-machine-learning recipe should transfer to non-cardiac fields such as oncology, neurology, and immunology, where mechanistic models are less mature; demonstrating benefit there would be a strong external confirmation.","The paper's own caveat that no data quantify M&S's contribution to rising life expectancy implies a concrete evaluative gap: a cost-effectiveness meta-analysis of in silico-guided versus conventional drug and device development would directly test whether the promised savings in time and money actually materialise.","An implicit, testable consequence is that classifiers trained on synthetic data should fail gracefully when a simulation omits a silent variable such as age; a benchmark that introduces such variables in held-out clinical sets would probe the limits of the whole synthetic-data strategy."],"forward_implications":["In silico trials built on cohorts of digital chimeras can reduce, refine, or replace some animal and early human experiments, and regulators are already beginning to accept simulation evidence.","Cardiac multi-scale models can supply large, well-labelled, class-balanced synthetic datasets that improve machine-learning classifiers beyond what scarce clinical data alone allow.","Physics-informed neural networks and statistical emulators will become routine tools for parameter identification, sensitivity analysis, and uncertainty quantification.","Interoperable standards and open, FAIR data and code are prerequisites for building and validating models across scales, diseases, and contexts of use.","The controlled in silico environment lets researchers isolate confounding factors and test cause-and-effect hypotheses in ways wet-lab experiments cannot easily match."],"supporting_citations":[{"why":"Sets out the millennial tools-and-visions agenda that the review revisits and against which it measures maturation.","marker":"[1]"},{"why":"Supplies the figure and concept of the digital twin as the vehicle for precision cardiology and for mechanistic/data-driven synergy.","marker":"[2]"},{"why":"Foundational proposal for integrating models from proteins to organs, grounding the whole-human modelling strand of the maturity narrative.","marker":"[3]"},{"why":"Source for the EU life-expectancy figures used as evidence of societal benefit, while noting that M&S's contribution is unquantified.","marker":"[14]"},{"why":"Defines in silico clinical trials, the methodological core of the virtual-cohort claims.","marker":"[29]"},{"why":"Provides the open synthetic ECG dataset that anchors the claim that simulated data can train machine-learning classifiers.","marker":"[39]"},{"why":"Defines the FAIR principles on which the open-data and open-tools argument depends.","marker":"[50]"},{"why":"Documents the regulatory guidance update that explicitly encourages in silico models to integrate ion-channel data, the strongest adoption example.","marker":"[132]"}],"fun_headline_variants":["Digital twins and chimeras: the new face of in silico medicine","Mechanistic meets machine learning: the future of medical simulation","AI and physics-based models: a natural fit for biomedicine","Simulation gets personal: digital twins and in silico trials","From one human model to many: cardiology's digital future"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the selected success stories—rising publication counts, digital-twin pilots, regulatory uptake, and synthetic-data machine-learning demos—are representative enough to prove that modelling and simulation has matured into an integral part of biomedical research, rather than being early or atypical exceptions.","fun_headline_variants_meta":{"raw":{"variants":["Digital twins and chimeras: the new face of in silico medicine","Mechanistic meets machine learning: the future of medical simulation","AI and physics-based models: a natural fit for biomedicine","Simulation gets personal: digital twins and in silico trials","From one human model to many: cardiology's digital future"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001949,"raw_usage":{"total_tokens":7663,"prompt_tokens":1031,"completion_tokens":6632,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":6542}},"tokens_in":647,"tokens_out":6632,"duration_ms":48027,"temperature":1.0,"reasoning_tokens":6542,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T16:37:40.611497+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A head-to-head comparison in which machine-learning classifiers trained on synthetic data from mechanistic cardiac models are outperformed, on out-of-sample clinical data, by classifiers trained on equal-sized clinical datasets for the same diagnostic task would undermine the synergy claim. So would a post-market audit showing that in silico trial predictions—for instance, proarrhythmia risk classifications—fail to correlate with adverse outcomes in subsequent real clinical trials.","supporting_citations":[{"cited_title":"2003 Integration from proteins to organs: the Physiome Project.Nature Reviews Molecular Cell Biology 4, 237–243","cited_arxiv_id":null,"evidence_quote":"Foundational proposal for integrating models from proteins to organs, grounding the whole-human modelling strand of the maturity narrative."}],"review_version":1}