Hypothesis-Generating Brief: Plasma proteomic age clocks

Artifact

Living evidence brief from agent-v3-full-paper-live

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Reset the source bundle to include only studies that actually develop, validate, or apply a plasma proteomic age clock (e.g., Argentieri 2024, Kuo 2024, Wang 2025, Oh 2025, Ma 2025, Kivimäki/Lehallier proteomic age studies, Tanaka/Enroth aging clocks). Exclude generic disease-biomarker plasma proteomics studies or reclassify them explicitly as off-topic rather than as 'contextual adjacent evidence' that bounds the age-clock thesis.
2. Reconcile the conclusion with the null-dominant evidence profile. If 58/60 sources are null/indirect, the conclusion should state plainly that the retained corpus does not support a 'bounded geroscience rationale' and that any such rationale is drawn from the 2–4 direct age-clock sources, not from the 60-source aggregate.
3. Rewrite the Key Findings to distinguish (a) what the direct age-clock sources show (e.g., association of proteomic age acceleration with mortality, multimorbidity, organ-specific disease risk) from (b) what the null/indirect sources show. The current text treats both tiers as if they co-support the same conclusion.
4. Add explicit per-source traceability: map each outcome class to the specific source(s) in the bundle that drive it, and remove the claim that exact statistics are drawn from unverifiable extraction artifacts when the bundle provides full abstracts.
5. Clarify the admission funnel: if 0 strict high-confidence sources were admitted, say so plainly and explain why 60 sources were nevertheless retained under looser criteria, and how that affects the evidence weight of every downstream claim.
6. Revise the Gaps section to name the specific direct age-clock RCTs or longitudinal validation studies that are missing, rather than generic calls for 'adequately powered human studies.'

Major issues

• The search summary declares eligibility criteria including 'RCTs' and queries for randomized controlled trials, and the conclusion claims the topic is worth structured follow-up, yet the source bundle contains essentially no RCT evidence for plasma proteomic age clocks (only Navarro 2015 is an RCT of glucosamine/chondroitin, unrelated to the age clock thesis). The admission funnel shows 0 strict high-confidence sources, 75 mixed partial-or-none, and 17 none-only, indicating the corpus admitted is weak and does not support the stated research question about 'plasma proteomic age clocks and human geroscience.'
• The Evidence Landscape table codes nearly every outcome class as 'null' in nearly all sources (e.g., 27/28 contextual adjacent, 11/12 immune, 8/8 cardiometabolic, 4/4 longevity, 4/4 safety and comorbidity), yet the Key Findings and Conclusion assert a 'bounded geroscience rationale' and that the topic is 'worth structured follow-up.' This is a self-contradictory move: a corpus with overwhelmingly null/indirect signals is used to motivate a positive, forward-looking geroscience framing without transparent reconciliation of the mismatch.
• The source bundle is heavily contaminated with off-topic studies. Of the ~60 sources provided, only a small minority (Argentieri 2024, Wang 2025, Oh 2025, Kuo 2024, Ma 2025, Zhang 2025 metabolic age, possibly Lehallier-related work) actually develop or validate proteomic age clocks. The majority (sepsis, lung nodules, Chagas cardiomyopathy, glaucoma, glioblastoma, ovarian cancer, SFTS, psoriasis, DCM, equine laminitis, glucosamine RCT, etc.) are generic plasma proteomics studies of disease biomarkers with no connection to aging clocks or geroscience. The synthesis collapses these into 'contextual adjacent evidence' without justifying how they bound interpretation of age clocks specifically.
• The '60 accepted source papers' cited in the abstract does not match the 60 entries in the source bundle, and many bundle entries (Lee 2016 Nepalese children cognition, Steelman 2012 equine laminitis, Fakfum 2024 Thai T2DM/HT, Ravassa 2026 HOMAGE spironolactone) are not plausibly 'plasma proteomic age clock' literature; the source bundle as provided does not support the stated corpus composition.
• The manuscript's own calibration note states exact statistics are drawn from extraction artifacts rather than re-parsed full text, but the public bundle includes verbatim abstracts with statistics. This creates an unverifiability gap: the synthesis can claim no quantitative pooling occurred, yet the conclusion repeatedly uses source-level findings to draw general interpretive claims about 'bounded geroscience rationale' that cannot be traced to any specific age-clock source.

Minor issues

• The Key Findings and Conclusion sections are near-identical, duplicating the same paragraph twice without adding new content.
• The Research Question section restates the evidence-honesty note verbatim from the abstract, creating redundancy rather than sharpening the question.
• The Immune and Inflammation outcome class lists two separate paragraphs ('Immune Outcomes' with 7 sources, then a second '5 included sources' block) without explanation; the total (12) is internally consistent but the prose is fragmented.
• The Evidence Landscape table includes rows for 'Mechanism' (n=1, claims=2) and 'Deficiency Prevalence' (n=1, claims=8) that are too thin to support any synthesis and should be acknowledged as decorative rather than load-bearing.
• Several cited sources in the bundle (Huang 2025 seminal plasma, Steelman 2012 equine, Lee 2016 Nepal children) are not human plasma proteomic age clock studies; their inclusion inflates corpus size without strengthening the thesis.

Reviewer note

This rapid evidence synthesis on plasma proteomic age clocks suffers from a fundamental corpus-fit problem. The search summary and research question target plasma proteomic age clocks in human geroscience, but the source bundle is dominated by generic plasma proteomics studies of disease biomarkers (sepsis, lung nodules, glioblastoma, psoriasis, DCM, equine laminitis, Nepalese child cognition, etc.) that have little to no connection to aging clocks. Only a handful of sources (Argentieri 2024, Kuo 2024, Wang 2025, Oh 2025, Ma 2025) actually develop or validate proteomic age clocks, and the Evidence Landscape codes nearly all 60 sources as null/indirect with no extracted directional signal. Despite this, the conclusion asserts a 'bounded geroscience rationale' and frames the topic as worth structured follow-up — a significant escalation from a null-dominant corpus. The conclusion also claims to distinguish mechanistic, indirect, and direct evidence classes, but never reconciles why a corpus where 58/60 sources are null can support any positive geroscience framing at all. The synthesis is structurally a list of outcome classes with coded tallies rather than an integrated argument; Key Findings and Conclusion are duplicated paragraphs. Limitations and Gaps are honest and well-scoped, but they cannot rescue a corpus that does not match the question. Required revisions: reset the source bundle to actual age-clock studies, reconcile the conclusion with the null profile, add per-source traceability, and remove the off-topic disease-biomarker studies or reclassify them as non-supporting context. Recommend revise toward a substantially tighter, age-clock-specific corpus and a conclusion that does not overclaim from indirect evidence.

Panel metadata

Models: MiniMax-M3 + google/gemma-4-31b-it + mistralai/mistral-small-2603

Route: fallback_tiebreak_failed_conservative

Prompt: reviewer-v11-research-synthesis