Image and Statistical Concerns in "Transcriptomic profiles link corticostriatal microarchitecture to genetics of neurodevelopment and neuropsychiatric risks" (Translational Psychiatry, 2025)

Academic fraud report · Geng Detector

Summary

This integrity review examines a 2025 article in Translational Psychiatry (DOI: 10.1038/s41398-025-03260-3) that integrates corticostriatal gene-expression data with HCP neuroimaging. The overall verdict is "questionable." The most robust concern is a statistical–interpretive inconsistency: the authors report an SVM classification accuracy of 72.2% for depression with a two-tailed P value of 0.065 (above their declared α = 0.05), yet include this non-significant result within the broad claim that corticostriatal circuits are "related to neuropsychiatric risks," constituting outcome-spin. A second confirmed issue involves methods-section language: the paper states that "all procedures… were approved by the local institutional review board and informed consent were obtained from all subjects," but the MRI data clearly come from the public HCP dataset, indicating template misuse or misleading ethics wording. Two further concerns remain unverified due to insufficient evidence: visually implausible pixel-level symmetry between Figure 4A and 4B PLS-1 scatter plots for left vs. right caudate, and overly smooth, parallel curves in Figure 1B and Figure 2B that lack expected biological noise. The report urges caution and recommends further data verification rather than definitive fraud claims.

Verdict

Questionable (🟡). Two concerns are confirmed (statistical spin; misapplied ethics template), while two image-related concerns are flagged but not substantiated.

Key findings

Statistical interpretation conflict (confirmed): Depression SVM accuracy 72.2% with two-tailed P = 0.065, above the declared significance threshold of P < 0.05, yet incorporated into the headline claim that corticostriatal circuits relate to neuropsychiatric risks.
Methods/ethics template misuse (confirmed): Paper claims "local IRB approval and informed consent from all subjects," but the MRI data are the public Human Connectome Project (HCP) dataset; the wording implies direct recruitment that did not occur.
Visually implausible left–right symmetry (unverified): Figure 4A and Figure 4B PLS-1 scatter plots, fitted curves, and error bars for Left caudate and Right caudate appear near pixel-level identical, including outlier positions—an outcome not expected from noisy transcriptomic/structural data.
Over-smoothed curves lacking biological noise (unverified): Figure 1B sub-panels and Figure 2B (LCAU, RCAU, LPUT, RPUT) show near-parallel trajectories with unusually uniform point spacing, suggesting aggressive smoothing or synthetic generation.

Evidence highlights

P = 0.065 vs. α = 0.05: A non-significant two-tailed result framed within a positive overall conclusion. Authors explicitly state two-tailed P < 0.05 as their significance criterion.
Dataset disclosure: The study reuses public HCP neuroimaging, yet the ethics statement references local IRB approval and informed consent—standard phrasing for primary data collection, not secondary analysis of a public repository.
Figure 4A / 4B: PLS-1 panels for Left caudate and Right caudate show curves and outlier locations that are visually indistinguishable, which is biologically implausible for independent hemispheric measurements.
Figure 1B / Figure 2B: Four cortical–striatal gradient curves display near-identical inflection points and parallel trends with minimal natural variability.

Notes

All quantitative values (P = 0.065, accuracy 72.2%, DOI 10.1038/s41398-025-03260-3) are preserved from the source report.
The image-symmetry and smoothness concerns are explicitly marked as "依据不足 / insufficient evidence" in the original; they warrant data and code requests but do not constitute proof of fabrication.
Recommended follow-up: request raw data points and plotting code for Figures 1, 2, and 4; raise PubPeer comments on the depression P-value interpretation and the suspicious left–right mirror quality.
This is an AI-assisted integrity screening for discussion only; final adjudication requires investigation by the journal or institution.