Integrity Review Report: Comparative transcriptome and genome analysis of susceptible Diantun 502 and resistant Diantun 506 rice upon Magnaporthe oryzae infection

Academic fraud report · Geng Detector

Summary

This review evaluates a 2025 BMC Plant Biology transcriptomic and genomic comparative study of two rice varieties, Diantun 502 (susceptible) and Diantun 506 (resistant), under Magnaporthe oryzae infection (DOI: 10.1186/s12870-025-06357-5). The overall verdict is 'Questionable' (yellow flag), indicating raised concerns but insufficient evidence for a definitive fraud determination. Two issues are flagged. First, the reported protein-protein interaction network claims an implausibly large topology (1,833 nodes and 11,966 edges) despite having only 3 biological replicates per condition (18 total samples) and PCA/clustering plots (Fig. 2A, 2B) showing highly dispersed D502 replicates at 24h and 48h. Re-review clarifies that the network is STRING-based PPI, not WGCNA, weakening but not eliminating the concern. Second, the 18 RNA-seq samples exhibit unusually homogeneous Q20 (95.48%–97.2%) and Q30 (89.07%–92.93%) values, consistent with default-parameter pipeline output. Both findings remain inconclusive. An originally alleged contradiction in Table 2 was invalidated upon re-reading the original text.

Verdict

Questionable (🟡). Concerns remain but fall short of conclusive evidence of misconduct.

Key findings

Implausibly precise PPI network topology with small/dispersed sample set: The authors report a STRING-based PPI network with 1,833 nodes and 11,966 edges for D502 at 24h, based on only 3 biological replicates per timepoint/condition (18 samples total). PCA (Fig. 2A) and clustering (Fig. 2B) plots indicate D502 replicates are highly dispersed at 24h and 48h, which makes the generation of such an exact and large topology unlikely. Note: re-review determined the network is STRING-derived (not WGCNA as originally alleged), substantially weakening the claim.
Unusually homogeneous RNA-seq QC metrics across all 18 samples: Table 1 reports Q20 values uniformly between 95.48%–97.2% and Q30 between 89.07%–92.93%, suggesting default-parameter pipeline processing without rigorous individualized quality assessment.

Evidence highlights

Sample size vs. network claim: 18 total samples (n=3 per condition/timepoint) yielding 1,833 nodes / 11,966 edges in PPI topology.
QC homogeneity: Q20 range = 1.72 percentage points (95.48%–97.2%); Q30 range = 3.86 percentage points (89.07%–92.93%) across all 18 libraries.
Replicate dispersion: Fig. 2A PCA and Fig. 2B clustering show D502 replicates highly scattered at 24h and 48h, yet precise topology parameters are still reported.
Original allegation retracted: The claim of a Table 2 contradiction (that authors erroneously attributed resistance-causing mutations to D506 instead of D502) was invalidated—the original text (p. 15) explicitly states "SNPs in this region might be responsible for inactivation of these genes in D502," and Table 2 correctly identifies amino acid variants in D502.
Image analysis limitation: No Western blot or microscopy figures with sufficient resolution were available for pixel-level PS artifact screening.

Notes

DOI: 10.1186/s12870-025-06357-5
Confidence: Low-to-moderate; both retained concerns rest on indirect indicators (sample-vs-output inconsistency, QC homogeneity) rather than direct fabrication evidence.
Recommended follow-up: (1) request raw sequencing data and original network-analysis parameters from authors, particularly regarding D502 replicate dispersion shown in Fig. 2; (2) post technical questions on PubPeer regarding pipeline rigor; (3) monitor the authors' subsequent publications for recurring patterns.
This report is AI-assisted and intended for academic discussion only; final determinations of misconduct require institutional investigation.