Investigation report: 'Identifying potential risk genes for clear cell renal cell carcinoma with deep reinforcement learning' (Nature Communications, DOI: 10.1038/s41467-025-58439-5)

Academic fraud report · Geng Detector

Summary

Verdict: Questionable (yellow). The Nature Communications 2025 paper (16:3591) by Lu, Zheng, Yi, Hao, Zeng, Han, Li, Jiao, Jiang, Ai & Peng was reviewed in text-only mode, which prevents pixel-level image forensics. Several serious concerns were nonetheless identified. (1) A clear textual contamination in the Results section: the fragment 'epointing out this issuefﬁcacies' is embedded mid-word in the shRNA-2/3 knockdown description, strongly suggesting paste-in of reviewer or draft text without proofreading. (2) Statistical methodology is internally inconsistent: Figure 4 legend specifies 'Two-tailed unpaired Student's t-test' while Methods states 'paired t-test', indicating copy-paste or mislabeling. (3) In vivo xenograft experiments use n=3 mice per group analyzed with Two-way ANOVA, which is widely regarded as underpowered for tumor growth studies. Image reuse in Figures 4–6 (Western blots, flow cytometry, tumor photos) could not be assessed. Confidence is moderate-to-high for the textual and statistical issues; image-integrity claims are explicitly out of scope pending access to original files.

Verdict

Questionable (🟡). Text-only review of the paper at DOI: 10.1038/s41467-025-58439-5 surfaced multiple non-trivial problems but did not yield conclusive evidence of data fabrication. The most defensible findings concern textual contamination and internal inconsistency in the statistical reporting.

Key findings

Textual contamination / apparent paste-in error. In the Results subsection on EGFR functional validation, the phrase describing shRNA-2/3 knockdown reads (verbatim): "short hairpin RNA (shRNA)-2/3 showed promising knockdown epointing out this issuefﬁcacies for EGFR silencing..." The fragment "pointing out this issue" has been spliced into the middle of "efficacies", producing the non-word "epointing out this issuefﬁcacies".
Internal contradiction in statistical test description. Figure 4 legend states: *"Two-tailed unpaired Student's t-test (L–O)"* (unpaired), whereas the Methods section *Statistical analysis* states: *"…determined using analysis of variance (ANOVA) and Student's paired t-test, respectively…"* The same type of in vitro comparison is described both ways.
Inadequate in vivo sample size. Figure 5B and 5C (xenograft assays) are annotated *"n=3 mice per group"*, and growth-curve analysis is reported using Two-way ANOVA. n=3 per group is widely considered underpowered for intergroup tumor-growth comparisons and is inconsistent with the computational rigor of the rest of the paper.
Image-integrity assessment not performed. Figures 4 (Western blots / flow plots), 5 (tumor images), and 6 (high-content imaging) were not assessable without the original high-resolution files; possible reuse of loading controls or tumor photographs cannot be ruled in or out from text alone.

Evidence highlights

DOI: 10.1038/s41467-025-58439-5
Journal: Nature Communications (2025) 16:3591
Verbatim textual defect: "…promising knockdown epointing out this issuefﬁcacies for EGFR silencing…"
Figure 4 legend (paraphrased): *Two-tailed unpaired Student's t-test (L–O).*
Methods (paraphrased): *…ANOVA and Student's paired t-test, respectively.*
Figure 5B/5C (paraphrased): *n=3 mice per group; Two-way ANOVA.*

Notes

All findings are derived from the provided text; pixel-level image forensics (loading-control reuse, cloning in tumor photographs, flow-plot duplication) require the original uncropped images and raw FCS files and could not be performed here.
The textual defect in the shRNA paragraph is most parsimoniously explained by an unrevised paste of a draft or reviewer response rather than intentional misconduct, but its presence in a published Nature Communications article indicates a failure of author proofreading, copy-editing, and peer review at the textual level.
The paired vs. unpaired t-test discrepancy is methodological: if the experiments were genuinely independent, a paired test is inappropriate; if the text is simply mis-copied Methods, it still indicates careless reuse of boilerplate.
Readers and integrity platforms (e.g., PubPeer) are encouraged to (a) request the original uncropped blots and tumor images, (b) seek clarification on the paired/unpaired test conflict, and (c) scrutinize the n=3 xenograft design.