Summary
This report flags Wang et al. (Nat. Cell Biol., DOI: 10.1038/s41556-026-01885-0) as highly suspicious but does not constitute proof of misconduct. The central quantitative concern is Figure 5h: with n=4 vs n=4, six independent cytokine comparisons yield P=0.029, 0.029, 0.029, 0.028, 0.027, and 0.029 by Mann–Whitney U test. The smallest attainable P-value for perfectly separated n=4 vs n=4 ranks is 0.0286 (≈0.029), so obtaining this exact bound across multiple, independent endpoints implies implausibly clean group separation. Secondary concerns include (i) implausibly short timeline between ethical approval B2024-759-01, sample collection in 67 locally advanced CRC patients with nICT follow-up, and submission on 1 August 2025; and (ii) unusually 'well-rounded' P-values in small mouse cohorts (e.g., P=0.0034, 0.0049, 0.0097, 0.0425, 0.0554), which warrant scrutiny but are not independently diagnostic. Pixel-level image analysis was not possible from the provided PDF/OCR text. Overall confidence in the anomaly pattern is moderate-to-high; confirmation requires raw data, blot/flow original files, and editorial investigation.
Verdict
🟠
Highly suspicious — Multiple statistical anomalies converge, but no single finding is independently conclusive. Confirmation requires raw data, original images, and an editorial/institutional investigation.
Key findings
- Figure 5h "0.029 paradox" (Severe): In an unpaired two-sided Mann–Whitney U test with n=4 vs n=4, the smallest attainable two-sided P-value is 1/(C(8,4)) = 1/70 ≈ 0.0143, but the smallest achievable value for completely non-overlapping groups is 2/(4!·4!) = 2/576 ≈ 0.00347; more relevantly, for the discrete permutation distribution the reported values 0.027, 0.028, 0.029 cluster near the minimum resolution boundary of the test, indicating multiple cytokines show group values that are perfectly or near-perfectly separated — biologically implausible across independent endpoints.
- P-values reported: P=0.029, 0.029, 0.029, 0.028, 0.027, 0.029 for six cytokines in Fig 5h.
- Suspiciously rounded P-values in mouse experiments (Moderate): Fig 2c P=0.0049; Fig 3b P=0.0034; Fig 3h P=0.0425; Fig 4k P=0.0097; Fig 7d P=0.0554. The pattern of P-values clustering at the 0.001 resolution near the 0.05 threshold in small samples (n≈5–6) is atypical.
- Timeline anomaly (Moderate): Ethics batch B2024-759-01 implies clinical collection began in 2024; manuscript received 1 August 2025. The study claims 67 nICT-treated locally advanced CRC patients (with pCR/cCR endpoints), 38 paired scRNA-seq, 5 snRNA-seq, conditional Cxcl12^fl/fl and Mdk^DTR mouse lines — a workload unlikely to be completed end-to-end within ~12 months.
- Image analysis: not feasible from supplied PDF/OCR; flagged as a follow-up requirement.
Evidence highlights
- Statistical floor argument (Fig 5h): With n=4 vs n=4 and tied/separated ranks, the permutation P-value distribution is coarse; multiple endpoints sitting at the minimum-resolution boundary (≈0.0286 rounded to 0.029) is the empirical fingerprint of non-overlapping data, which is statistically improbable across biologically distinct cytokine measurements.
- Small-sample P-value clustering: n=5–6 mouse cohorts typically produce irregular P-value distributions; the observed clustering (0.0034, 0.0049, 0.0097, 0.0425, 0.0554) suggests possible P-value selection or post-hoc trimming.
- Timeline vs. workload: 67-patient nICT cohort with response endpoints + multi-omic profiling + multi-line mouse genetics + mechanistic rescue experiments between 2024 ethics approval and 1 August 2025 submission is difficult to reconcile without pre-existing data or a very large team.
- Preserved identifiers: DOI 10.1038/s41556-026-01885-0; ethics ID B2024-759-01; submission date 1 August 2025; sample sizes n=4, n=5, n=6, n=38, n=5 (snRNA-seq), 67 (cohort).
Notes
- Pixel-level image forensics (Western blots, flow cytometry scatter, microscopy) could not be performed on the supplied PDF text/OCR; raw .tif/.czi/.fcs files and high-resolution figures must be requested from the authors.
- Mann–Whitney U is reported here as the test used; if the authors instead used an unpaired t-test on n=4 vs n=4, the minimum P-value for fully separated data is the same order (≈0.0035 two-sided), but 0.029 still requires substantial separation across all six cytokines — the anomaly stands under either test.
- The "Geng" style framing and the 0.029 bound argument should be treated as a flag, not a verdict. A definitive determination requires: (a) individual data points for Fig 5h; (b) uncropped blot and flow files; (c) patient enrolment and follow-up logs against ethics approval B2024-759-01.
- This report is AI-assisted and intended for academic discussion only.
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/report/geng_geng_6a29938725f6d8.40357599