Integrity Assessment Report: Identifying potential risk genes for clear cell renal cell carcinoma with deep reinforcement learning (Nature Communications, 2025)

Academic fraud report · Geng Detector

Summary

This report examines a Nature Communications article (DOI: 10.1038/s41467-025-58439-5) on deep reinforcement learning-based risk gene identification for clear cell renal cell carcinoma (ccRCC). The overall verdict is highly suspicious. The most severe textual finding is a copy-paste artifact in the Figure 4 caption where reviewer language ('pointing out this issue') is concatenated with experimental text, producing the non-word 'epointing out this issuefficacies'. Additional concerns include implausibly small sample sizes (n=3 mice per group in Figure 5, n=3–4 in vitro) coupled with claims of robust statistical significance, inconsistent sample-size reporting (n=3 in some methods sections vs. n=5 independent experiments in Figure 4), a duplicated reference to the same Nature Genetics paper differing only in author abbreviation format, and a mismatched citation justifying a tumor-volume formula. The image pixel-level analysis was not performed. Confidence in the textual evidence is high, but final judgment requires editorial investigation and raw data verification.

Verdict

🟠 Highly suspicious. The most damning textual artifact (embedded reviewer comment in the Figure 4 caption) is almost certainly not a typographical error and points to careless or rushed manuscript handling. Combined with statistical descriptions inconsistent with accepted norms for in vivo studies, the paper warrants formal editorial investigation. Image-level manipulation cannot be ruled out from text alone and requires independent examination of the original blots, microscopy, and flow cytometry data.

Key findings

🔴 Embedded reviewer text in main text (Figure 4 caption): A fragment of reviewer language was pasted into the experimental description, producing the non-word "epointing out this issuefficacies" — a concatenation of reviewer phrasing "pointing out this issue" with the original word "efficacies". This is a clear copy-paste artifact inconsistent with professional manuscript handling.
🔴 Implausibly small sample sizes vs. claimed significance: In vivo experiments in Figure 5 reportedly use n=3 mice per group, with in vitro assays at n=3 or n=4. Robust statistical significance in tumor-growth inhibition studies with such sample sizes is unusual and suggests cherry-picked or fabricated data.
🔴 Inconsistent sample-size reporting: Methods state experiments were "performed in triplicate", yet Figure 4 reportedly describes n=5 independent experiments — an internal contradiction that undermines data credibility.
🟡 Duplicate reference: The same Nature Genetics paper (Leiserson et al., *Pan-cancer network analysis…*, Nat. Genet. 47, 106–114, 2015) appears in the reference list twice — Ref 30 and Ref 37 — differing only in author-abbreviation format.
🟡 Mismatched methodological citation: A standard tumor-volume formula ($L \times W^2 \times 0.5236$) is supported by a citation to a prostate cancer Vinculin study (Zheng, X. et al., *Prostate* 81, 347–356, 2021), suggesting a patchwork methodology section rather than a coherent justification.
🟡 Compressed experimental timeline: Animal ethics approval No. 20230214007 (14 February 2023) and submission on 13 May 2024 leave roughly 15 months for animal approval, multi-gene (EGFR, PCLO) stable-line construction in multiple cell lines (ACHN, 786-O, 293T), and a full battery of assays (CCK-8, apoptosis, migration, invasion, xenografts) — a suspiciously short production cycle.

Evidence highlights

Exact verbatim artifact from Figure 4 caption: *"short hairpin RNA (shRNA)-2/3 showed promising knockdown epointing out this issuefﬁcacies for EGFR silencing"* — the italicized portion is non-English and combines a reviewer comment with the original word "efficacies" (with the ligature ﬁ preserved).
Stated sample sizes referenced in the report: n=3 mice per group (Figure 5 in vivo), n=3–4 (in vitro), and n=5 independent experiments cited in Figure 4.
Duplicate reference identifiers: Ref 30 and Ref 37, both pointing to Leiserson et al., *Nat. Genet.* 47, 106–114 (2015).
Cited tumor-volume formula: $L \times W^2 \times 0.5236$, supported by Ref 117 (Zheng, X. et al., *Prostate* 81, 347–356, 2021).
Timeline anchors: ethics approval 20230214007 (14 February 2023); Received 13 May 2024; Accepted 18 March 2025.
DOI: 10.1038/s41467-025-58439-5.

Notes

This assessment is constrained to textual, statistical-description, citation, and metadata analysis. No pixel-level examination of Western blots, microscopy images, or flow cytometry plots was performed; such image-based analysis is necessary to fully evaluate the figures.
The "epointing out this issuefficacies" artifact is, on its own, strong evidence of unprofessional manuscript preparation. Whether it reflects fraud, extreme carelessness, or a language-processing failure is for the editorial office to determine.
The small in vivo sample size (n=3) is unconventional but not by itself disqualifying; however, when combined with claimed strong statistical significance and contradictory n=5 reporting elsewhere, it substantially weakens the evidentiary weight of the wet-lab claims.
The report's recommendation is to request the corresponding authors provide raw unprocessed Western blot images, original flow cytometry FCS files, complete animal husbandry records, and full individual-animal tumor-volume measurements, and to file a PubPeer comment citing the verbatim Figure 4 caption artifact.
Final determination of academic misconduct must be made by a qualified institutional or editorial investigation, not by this report.