{"id":"0d43e34b-b9d6-4d95-8bab-b7cabf6aa344","arxiv_id":"2508.13354","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A multi-scale epidemic model shows that social contacts shorten the generation time and increase pre-symptomatic transmission compared to a purely biological baseline.","lead":"VIBES is a new multi-scale model that combines within-host viral dynamics with a social contact network to simulate epidemic spread. It provides quantitative estimates of generation time and pre-symptomatic transmission for SARS-CoV-2, disentangling biological from social drivers.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Baseline for network effect is not a no-network epidemic null; realized vs intrinsic generation time confound.","rationale":"The reader's weakest assumption concerned the unknown mapping from viral load to infectiousness, which is indeed a critical precondition for any numerical claim. However, even if that mapping is correct, the abstract's comparison cannot justify the stated effect of the social network because the 'biological baseline' is not an epidemic null. This is a structural flaw in the causal attribution, more directly tied to the strongest claim. I therefore partially agree: the infectiousness mapping is a legitimate concern, but the more load-bearing issue is the invalid baseline comparison. My proposed test would settle whether the network specifically matters. Since the full text might contain a well-mixed control that the abstract omitted, I do not reject outright; I recommend a conditional verdict: the central claim should be revised or supported by the missing control analysis. If the well-mixed control reproduces the full-model results, the claim that social networks shorten generation time and increase pre-symptomatic transmission is false and should be rejected. If not, the claim stands. The current abstract does not provide this evidence, so the verdict should shift from UNVERDICTED to CONDITIONAL pending this check.","tokens_in":805,"tokens_out":4953,"duration_ms":57514,"concrete_test":"Implement a well-mixed (homogeneous-mixing) version of the full VIBES model using the same within-host infectiousness profile, same R=3.0, same fraction symptomatic/asymptomatic, and same isolation behavior, but with random contacts instead of the empirical network. Compute the mean generation time and pre-symptomatic transmission. If these match the network-model values (5.4 d, 52.8%) within confidence intervals, then the social network contributes no additional effect beyond transmission dynamics, and the central attribution to social drivers fails. If they differ significantly, the network effect is supported. This test separates the network's contribution from intrinsic-vs-realized generation interval bias and from symptom-status composition effects.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim—that adding a realistic social contact network shortens generation time from 6.3 to 5.4 days and raises pre-symptomatic transmission from 43.1% to 52.8%—rests on a comparison against a 'purely biological baseline' that is not a valid no-network counterfactual. That baseline is computed from the within-host model alone, yielding the intrinsic generation interval and intrinsic pre-symptomatic fraction of the infectiousness profile. The full model, by contrast, simulates an epidemic with multiple infected individuals, competition for susceptible contacts, and a realized set of transmission events. Even in a well-mixed epidemic, the observed (realized) generation interval is known to be shorter than the intrinsic one when the epidemic is growing, because faster-transmitting branches are preferentially observed. The 0.9-day difference may therefore reflect this intrinsic-vs-realized bias rather than any effect of network structure. Additionally, the baseline is reported for symptomatic individuals only, while the full-model generation time likely pools symptomatic and asymptomatic individuals, further confounding the comparison. The abstract's 'disentangling' claim is thus unsupported: the design does not isolate the social-contact network from other differences between a within-host profile and a full epidemic process. A proper null would be a well-mixed epidemic model with the same within-host infectiousness, same R, same symptom-status distribution, and same interventions.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces VIBES, a multi-scale modeling framework that couples within-host viral dynamics (fitted to patient-level data) with between-host transmission on a data-driven social contact network. Using SARS-CoV-2 as a case study, the authors report a purely biological baseline generation time of 6.3 days and 43.1% pre-symptomatic transmission for symptomatic individuals. Adding the social-contact network shortens the generation time to 5.4 days and raises pre-symptomatic transmission to 52.8% at R=3.0. They further report that increasing transmissibility (R from 1.3 to 6) shortens generation time and serial interval by up to 21% and 13%, that isolation increases the pre-symptomatic fraction by about 30%, and that the framework yields an asymptomatic generation time of 5.6 days at R=1.3. The central claim is that this framework disentangles biological from social drivers of epidemic dynamics.","tokens_in":1151,"tokens_out":2475,"duration_ms":27717,"significance":"If the methodology is sound, VIBES would be a valuable contribution: it combines patient-level viral dynamics with network structure and produces quantitative, mechanistically interpretable outputs that are difficult to obtain empirically. The reported estimates (generation time, serial interval, pre-symptomatic fraction) are directly relevant to epidemic modeling and public health intervention design. The use of a patient-data-driven within-host component and a data-driven contact network are notable strengths. However, because this review is based on the abstract only, I cannot verify the fitting procedures, parameter identifiability, sensitivity analyses, or the precise definition of the biological baseline. The central attribution claim—that the network is responsible for the shortening of the generation interval—depends on a comparison that may conflate intrinsic and realized generation times, as detailed below. The framework's potential is real, but the evidence in the abstract does not yet establish the causal disentangling claimed.","major_comments":[{"comment":"The baseline for the network effect is the intrinsic generation interval from the within-host model, not a no-network epidemic null. In a growing epidemic, realized generation times are systematically shorter than intrinsic ones because faster-transmitting branches are preferentially observed even in a well-mixed model. The reported shortening from 6.3 to 5.4 days may therefore reflect this intrinsic-versus-realized bias rather than any effect of network structure. A proper null is a well-mixed epidemic model with the same within-host infectiousness profile, the same R, and the same symptom-status distribution as the network model. The authors should run this control and report whether the network effect persists.","section":"Abstract"},{"comment":"The biological baseline is reported for symptomatic individuals only, whereas the full model's generation time and pre-symptomatic fraction appear to pool symptomatic and asymptomatic individuals. Because the authors estimate an asymptomatic generation time of 5.6 days, inclusion of asymptomatic cases in the full model can itself shorten the pooled generation time relative to a symptomatic-only baseline. To support the attribution claim, the authors should report both symptomatic-only and combined baselines for the well-mixed control and for the network model.","section":"Abstract"},{"comment":"The mapping from within-host viral load to infectiousness is not described. The statement that the baseline is 'from the within-host model' leaves unspecified whether infectiousness is assumed proportional to viral load, whether host-level variation (e.g., superspreading) is included, and whether the viral-kinetic parameters are fitted or fixed. This mapping is load-bearing: any error in it propagates to all emergent metrics (generation time, serial interval, pre-symptomatic fraction). The full text should provide the functional form, parameter values, and any validation of this mapping against independent transmission data.","section":"Abstract"}],"minor_comments":[{"comment":"The phrase 'biological baseline, thus independent of the reproduction number (R)' is confusing because the full model is evaluated at specific R values. Clarify how R is imposed in each model and why the baseline is independent of it.","section":"Abstract"},{"comment":"The term 'disentangling' overstates causal attribution given the comparison issues. Consider using 'quantifying the contribution' or 'decomposing the effects' unless the null-model analysis fully supports the causal claim.","section":"Abstract"},{"comment":"Definitions of generation time, serial interval, and pre-symptomatic transmission should be stated precisely. In particular, specify the infectious-period threshold used to classify pre-symptomatic versus symptomatic transmission.","section":"Abstract"},{"comment":"No data or code availability statement is visible in the abstract. If the full paper includes code for the VIBES framework and the data-processing pipeline, this should be highlighted.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"This review was conducted on the abstract only because the full text was not supplied. My recommendation reflects a load-bearing concern about the comparison that underlies the central 'disentangling' claim: the intrinsic-versus-realized generation interval confound may invalidate the headline result unless a well-mixed null is added. I cannot assess the statistical fitting or the sensitivity of results to the viral-load-to-infectiousness mapping without the methods. The manuscript should be sent back for a full review once these issues are addressed; the framework itself is potentially valuable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a plausible framework with a real gap in the headline comparison. I'd send it to a referee, but the disentangling claim as stated is not yet supported.\n\nThe genuinely new thing is the explicit integration of a within-host viral dynamic model fit to patient data with a data-driven contact network, used to derive emergent metrics like the generation time for asymptomatic cases that are hard to measure empirically. That is a useful contribution. The authors are appropriately specific about what the baseline does and does not include: the biological baseline is independent of R and derived from the within-host model. The concrete numbers—6.3 vs 5.4 days and 43.1% vs 52.8%—will be useful benchmarks for other modelers if they hold up.\n\nThe soft spot is exactly the one the stress test flags. The 'purely biological baseline' is the intrinsic generation interval of the infectiousness profile. The full model produces a realized generation time in a simulated epidemic. In any growing epidemic, realized intervals are shorter than intrinsic ones because fast-transmitting branches are preferentially counted. So the 0.9-day shortening may reflect growth bias rather than network structure. The baseline also appears limited to symptomatic individuals, while the full model likely pools all cases. The causal attribution of the shortening to social contact structure is therefore not established by the numbers in the abstract. It would be if the authors compared against a well-mixed epidemic model with the same within-host infectiousness, same R, and same symptom-status mix. If the full text contains that comparison, my concern goes away—but the abstract does not report it.\n\nA smaller issue: infectiousness is presumably mapped from viral load, and the abstract does not state whether that mapping is calibrated or whether host-level variation is included. Standard modeling choice, but it makes the 'biological' baseline less biological than it sounds.\n\nBottom line: worth a careful read. The idea is solid and the estimates are concrete, but the central claim about disentangling drivers needs a better control. I'd accept it for peer review, specifically to force the comparison against a well-mixed null.","headline":"Plausible multi-scale framework, but the headline 'disentangling' claim compares intrinsic to realized generation time and needs a well-mixed null.","tokens_in":1580,"tokens_out":2569,"would_cite":false,"duration_ms":26299,"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 multi-scale model disentangles biological from social drivers of epidemic timing, showing social contacts shorten generation time and boost pre-symptomatic spread.","keywords":["multi-scale modeling","within-host dynamics","between-host transmission","generation time","serial interval","pre-symptomatic transmission","social contact network","SARS-CoV-2"],"falsifier":"A direct test would be to compare the model's predicted R-dependent shortening of generation time against empirical contact tracing data that records both serial intervals and local R estimates for SARS-CoV-2; if no shortening appears when R rises, the social-competition mechanism is wrong.","tokens_in":784,"feed_emoji":"🦠","tokens_out":1911,"duration_ms":21728,"temperature":0.7,"pith_summary":"This paper introduces VIBES, a multi-scale modeling framework that integrates within-host viral dynamics (from patient-level data) with between-host transmission on a data-driven social contact network. Using SARS-CoV-2 as a case study, the authors aim to separate biological drivers from social drivers of three epidemic properties: generation time, serial interval, and the proportion of pre-symptomatic transmission. They first establish a purely biological baseline, independent of the reproduction number, finding a generation time of 6.3 days and 43.1% pre-symptomatic transmission. Adding the social contact network at R=3.0 shortens the generation time to 5.4 days and raises pre-symptomatic transmission to 52.8%, demonstrating that social structure measurably alters key outbreak metrics. The framework also lets them estimate otherwise hard-to-measure quantities, such as an asymptomatic generation time of 5.6 days.","feed_headline":"Social contacts shorten COVID generation time to 5.4 days","feed_subtitle":"Model links viral dynamics to human networks, raising pre-symptomatic spread to 52.8%.","key_machinery":"The central machinery is the VIBES framework—a multi-scale model that couples a within-host viral dynamics model, calibrated to patient-level data, with a between-host transmission model on a data-driven social contact network. The within-host component provides a purely biological baseline for infectiousness over time, independent of R, while the network component adds realistic human contact structure whose effect depends on R. The framework's key output is the emergent timing metrics (generation time, serial interval, and pre-symptomatic transmission proportion), which are compared between the biological-only baseline and the full model to isolate the social driver.","core_discovery":"VIBES claims to mechanistically quantify how pathogen biology and human social behavior jointly shape epidemic dynamics. The central discovery is that a purely within-host biological baseline, which does not depend on the reproduction number, yields a generation time of 6.3 days for symptomatic individuals and 43.1% pre-symptomatic transmission. When a data-driven social contact network is added, the generation time shortens to 5.4 days and pre-symptomatic transmission increases to 52.8% at R=3.0. The paper further shows that as transmissibility rises (R from 1.3 to 6), competition among infectious individuals shortens generation time and serial interval by up to 21% and 13%, respectively, w","pith_inferences":["The same framework could be ported to other respiratory pathogens with available within-host viral load data and contact network data, offering a general way to separate biological from behavioral drivers.","A testable prediction arises: in settings with higher R, observed serial intervals should be systematically shorter—empirically checkable with contact tracing datasets that record R or secondary attack rates.","The reported increase in pre-symptomatic transmission under isolation suggests that isolation policies may inadvertently select for transmission during the pre-symptomatic window, a tradeoff the authors mention but do not fully explore.","If the within-host-to-infectiousness mapping varies by host (e.g., due to immunity or variant), the baseline estimates would shift; this is an untested assumption that the framework will need to incorporate."],"forward_implications":["If VIBES is correct, public health forecasts that use a single fixed generation time or serial interval are missing a social-structure effect that can shift these values by roughly 15% or more.","Interventions like isolation do not just reduce overall transmission; they change the route mix of transmission, increasing the share of pre-symptomatic spread by about 30%, which should inform contact tracing priorities.","The framework provides a way to estimate generation times for asymptomatic individuals—5.6 days at R=1.3—which are difficult to obtain empirically and are often excluded from outbreak models.","As pathogen transmissibility rises, the generation time and serial interval shorten (up to 21% and 13% respectively), meaning epidemic response timelines should adjust to the current R level."],"supporting_citations":[],"fun_headline_variants":["Social networks cut COVID generation time to 5.4 days from 6.3","Pre-symptomatic COVID spread hits 52.8% with social contacts modeled","How social mixing accelerates COVID: generation time shrinks to 5.4 days","VIBES model: social contacts shorten COVID generation time to 5.4 days","COVID generation time drops 14% with social contact networks"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The mapping from within-host viral load to infectiousness is assumed to be a deterministic proxy for transmission probability, and the entire biological baseline inherits the accuracy of that mapping.","fun_headline_variants_meta":{"raw":{"variants":["Social networks cut COVID generation time to 5.4 days from 6.3","Pre-symptomatic COVID spread hits 52.8% with social contacts modeled","How social mixing accelerates COVID: generation time shrinks to 5.4 days","VIBES model: social contacts shorten COVID generation time to 5.4 days","COVID generation time drops 14% with social contact networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00109,"raw_usage":{"total_tokens":4457,"prompt_tokens":877,"completion_tokens":3580,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":621,"completion_tokens_details":{"reasoning_tokens":3491}},"tokens_in":621,"tokens_out":3580,"duration_ms":26018,"temperature":1.0,"reasoning_tokens":3491,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T19:05:00.909339+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test would be to compare the model's predicted R-dependent shortening of generation time against empirical contact tracing data that records both serial intervals and local R estimates for SARS-CoV-2; if no shortening appears when R rises, the social-competition mechanism is wrong.","supporting_citations":[],"review_version":1}