Geng Academic Fraud Report: CDCAs Methylation and Immune Infiltrates in Hepatocellular Carcinoma (Frontiers in Oncology, 2021)

Academic fraud report · Geng Detector

Summary

This report evaluates a 2021 Frontiers in Oncology paper (DOI: 10.3389/fonc.2020.566183) that performs a bioinformatics analysis of CDCAs in hepatocellular carcinoma. Verdict: highly suspicious (orange tier). Three concrete findings were confirmed: (1) a statistical impossibility in Table 1, where CDCA5's multivariate Cox model reports HR=0.721 with 95% CI 0.508–1.023 (crossing 1) yet P=0.049, which violates basic statistical logic, and CDCA4 shows identical P-values (0.030) in univariate and multivariate models; (2) a template artifact in the Author Contributions stating that contributors performed 'experiments', despite the study being purely computational; (3) an internal contradiction, where the abstract reports seven over-expressed CDCAs (excluding CDCA4) while the Results section claims eight. Image-level duplication could not be assessed from the available text. Overall confidence is moderate-to-high for findings 1–3, with explicit limits on image-based detection. Final determination requires institutional investigation.

Verdict

🟠 Highly suspicious. Three confirmed findings indicate data fabrication, template-based writing, and internal inconsistency, though no image-level manipulation could be verified from textual evidence alone.

Key findings

Statistical impossibility in Cox regression (Table 1, Page 10): CDCA5 reported with HR=0.721, 95% CI=0.508–1.023, P=0.0490. Because the CI upper bound exceeds 1, P must be ≥0.05; the reported P<0.05 is mathematically impossible from any legitimate regression output. CDCA4 shows identical P=0.030 in univariate and multivariate analyses, which is implausible after adjusting for Age and Gender.
Author Contributions template artifact (Page 12): Statement that contributors "contributed to the experiments of the manuscript" is inconsistent with a purely computational study using R packages (limma, WGCNA) and online tools (TIMER, DiseaseMeth). No wet-lab experiments are described.
Internal contradiction on CDCA4 expression: Abstract states seven CDCAs (CDCA1–3 and CDCA5–8) are over-expressed in HCC; Figure 3 and Results claim all eight CDCAs are over-expressed. The discrepancy signals sloppiness or splicing of heterogeneous analyses.

Evidence highlights

Table 1, multivariate row for CDCA5: HR 0.721 (95% CI 0.508–1.023), P = 4.90E-02. CI crosses 1 ⇒ P must be ≥0.05. Reported P violates this constraint.
Table 1, CDCA4 univariate P = 3.00E-02; multivariate P = 3.00E-02. Identical values across models including Age and Gender are statistically implausible.
Author Contributions: "YW, assisted by HG, TO, LZ, JH, and SH, contributed to the experiments of the manuscript." No experimental procedures are documented anywhere in the paper.
Abstract vs. Figure 3 caption: Abstract → 7 CDCAs over-expressed (CDCA1–3, CDCA5–8); Figure 3 → 8 CDCAs over-expressed.
Image-based checks (Figures 1–10): insufficient raw resolution data provided; could not assess duplication or splicing.

Notes

DOI preserved as 10.3389/fonc.2020.566183; all numeric values reproduced verbatim from the source report.
Confidence: high for Finding 1 (mathematically deterministic); moderate for Findings 2 and 3 (textual evidence is unambiguous but motive is inferred); low for image-level concerns (untestable from supplied material).
Recommended actions: request raw R session output for the Cox regressions from the authors; raise concerns on PubPeer citing the CI/P inconsistency and template artifact; notify the journal editorial office to verify data authenticity.
Limitations: pixel-level image analysis, raw TCGA data reconciliation, and full Figure 3 quantitative reanalysis were not performed. Final determination of misconduct requires institutional investigation.