Fraud Investigation Report: CGMega (Nature Communications, 2024) – Suspect antibody catalog mismatches and implausible n=3 statistics

Academic fraud report · Geng Detector

Summary

Verdict: highly suspicious. The dry-lab (bioinformatics) component of this Nature Communications paper appears internally coherent, but the wet-lab validation section displays fabricated-looking experimental details. The most damning evidence is in the Methods: the BRCA2 antibody is described as Abcam clone EPR23442-43, yet assigned Proteintech catalog number 29450-1-AP; the ROCK2 antibody is attributed to Santa Cruz with sc-100425, yet also given Proteintech catalog 21645-1-AP. Misplacing a vendor catalog in this manner is uncharacteristic of a real experimenter and resembles AI-hallucinated or copy-pasted text. A second concern is statistical implausibility: Figure 4h reports P = 0.0023 from a paired two-tailed t test with n = 3 biological replicates (df = 2), implying an effect size roughly 15.8 standard errors above the mean – an extraordinarily unlikely outcome in real cell-viability assays. Pixel-level image analysis was not possible because source images were not provided. Confidence is high on the catalog mismatches (textual, verifiable) and moderate on the statistical anomaly; image manipulation cannot be confirmed or excluded without raw Western blots and plate-reader files.

Verdict

🟠 Highly suspicious. Textual evidence of fabricated or AI-generated antibody metadata, combined with a statistically implausible result from a minimal sample size, raises serious concerns about the integrity of the wet-lab validation in this paper.

Key findings

Antibody catalog/clone mismatch (Methods, Page 12, Western blot analysis): BRCA2 is described as Abcam clone EPR23442-43 but given Proteintech catalog 29450-1-AP; ROCK2 is attributed to Santa Cruz with sc-100425 yet assigned Proteintech catalog 21645-1-AP. The -AP suffix is characteristic of Proteintech rabbit polyclonal products and is inconsistent with the stated vendor and clone information. Apparent gene aliases (FACD/FANCD1, KIAA0619) appear in place of real clone identifiers.
Statistically implausible p-value at n = 3 (Figure 4h): A paired, two-tailed t test with n = 3 biologically independent experiments (df = 2) yields P = 0.0023. The implied t-statistic (~15.8) requires the mean of paired differences to exceed the standard deviation by a factor of roughly 10, an outcome essentially never observed in genuine cell-viability assays.
No image-level analysis possible: Western blot panels (Figure 4e) and cell-viability bar/scatter plots (Figures 4f, 4g) were not available for pixel-level inspection; therefore splicing, background manipulation, or duplicated bands cannot be confirmed or excluded.

Evidence highlights

DOI: 10.1038/s41467-024-50426-6 (Nature Communications, accepted 9 July 2024).
BRCA2: claimed vendor = Abcam, clone = EPR23442-43, but catalog = 29450-1-AP (Proteintech product line).
ROCK2: claimed vendor = Santa Cruz, catalog = sc-100425, but assigned catalog = 21645-1-AP (Proteintech product line).
Figure 4h legend states n = 3 biologically independent experiments; reported P value = 0.0023, paired t test, two-tailed, for olaparib + RKI-1447 combination vs. monotherapy.
Supplementary Source Data and full-membrane Western blot images were not provided to the reviewer and are required for confirmation.

Notes

Confidence is high for the antibody metadata errors because they are verifiable from the published text alone.
Confidence is moderate for the statistical anomaly: while the calculation is straightforward, an unusually large but genuine biological effect cannot be formally excluded without the raw plate-reader data.
No pixel-level conclusion can be drawn regarding Figures 4e–g until the original uncropped Western blot membranes and raw cell-viability data are released.
Recommended follow-up: request full uncropped Western blot membranes, raw plate-reader exports for the n = 3 replicates, and a PubPeer comment focusing on the catalog/clone inconsistencies and the n = 3, P = 0.0023 result.