Integrity Assessment Report: Comprehensive single-cell transcriptomic atlas of microglia in Alzheimer's disease mouse models

Academic fraud report · Geng Detector

Summary

Verdict: cleared (no substantive integrity issues detected). This report evaluates the Molecular Psychiatry article by Ping Liu et al. (2026, DOI 10.1038/s41380-026-03529-z), a single-cell transcriptomic atlas of microglia in Alzheimer's disease mouse models. The assessment was conducted as a text- and metadata-level review; pixel-level forensic image analysis was not performed because the original high-resolution figures were not available for inspection. Statistical and methodological checks found no irregularities: reported R² values span a realistic range across models and phenotypes (e.g., 0.666 in Figure 5 for plaque load vs. Tmem119+ area; 0.051 for APPNL-G-F and 0.333 for 5xFAD correlations with Tmem119 in Figure 7), with no suspiciously 'too-perfect' dose-response relationships. The authors explicitly state that mouse-level means were used as the unit of analysis, reducing pseudoreplication risk. The publication timeline (IACUC approval 2024-07-10, acceptance 2025-12-22, publication 2026-03) is internally consistent, software versions (R 4.3.1, Seurat 5.3.0) are appropriate for late-2025 analyses, and all cited references precede publication. Limitations: absence of pixel-level image inspection means image duplication or splicing cannot be ruled out.

Verdict

Cleared. No substantive integrity issues were identified at the text and metadata level. Image-level forensic analysis was not possible due to absence of original high-resolution figure files.

Key findings

Statistical reporting shows realistic biological variability: R² values range widely across models and outcomes (Figure 5: R² = 0.666 for plaque load vs. Tmem119+ area; Figure 7: R² = 0.051 for APPNL-G-F and R² = 0.333 for 5xFAD), with no suspiciously perfect correlations.
Age-dependent amyloid pathology in 5xFAD mice (Figure 8: 1.5, 3, 6, 12 months) follows the expected biological trajectory—no plaques or expression at 1.5 months, progressive accumulation with age.
Appropriate statistical methods were used (t-tests for two groups, ANOVA with Holm-Šídák or Tukey's post-hoc for multiple groups), and the Methods explicitly declare use of mouse-level means as the unit of analysis, a rigorous choice that reduces false-positive risk.
Experimental design is well-documented (5–6 mice per experimental group; 6–10 fluorescence images acquired from 2–3 fixed tissue sections), with no internal contradictions in the text.
Timeline is internally consistent: IACUC approval 2024-07-10, experiments completed before acceptance 2025-12-22, published 2026-03.
Software stack is appropriate: R v4.3.1, Seurat v5.3.0, with hdWGCNA citing recent literature.
All cited references (including 2025 Lancet and Cell reviews on Aβ-targeted therapies, and 2023–2025 single-cell references) predate submission; no anachronistic citations were found.

Evidence highlights

DOI: 10.1038/s41380-026-03529-z
Journal: Molecular Psychiatry (2026) 31:4164–4180
Corresponding authors: Peng Zheng, Ma-Li Wong
Reported R² values: 0.666 (Fig. 5, plaque load vs. Tmem119+ area); 0.051 (Fig. 7, APPNL-G-F); 0.333 (Fig. 7, 5xFAD)
IACUC approval number: IACUC-CQMU-2024-0710
Acceptance date: 2025-12-22; Publication: 2026-03
Sample sizes stated in Methods: 5–6 mice per group; 6–10 fluorescence images from 2–3 tissue sections per animal
Statistical unit of analysis: mouse-level mean (explicitly declared in Methods)

Notes

This review was limited to text-level and metadata analysis; pixel-level forensic examination of figures (for duplication, splicing, or background anomalies) was not conducted because original high-resolution image data was not available.
The absence of text- or metadata-level anomalies does not categorically exclude image manipulation; original data and raw images would be required for a definitive forensic conclusion.
Authors' explicit use of mouse-level means as the statistical unit is a notable methodological strength that reduces pseudoreplication, which is a common pitfall in studies with multiple images per animal.