Review Papers Claim by Claim

 Quick Explanation Copied

 Long Explanation

Scope and attribution caveat. The query names Shu Yuan, but none of the six supplied evidence records include author lists, so Shu Yuan's authorship of any specific paper cannot be verified from the provided data. This review therefore evaluates the six supplied papers as the evidence corpus, spanning pharmacovigilance machine learning, human proteomics, livestock genomics, forestry physiology, aquaculture nutrition, and spatial-omics methods.

Computational benchmarks: real but narrow gains

DSEP trained a back-propagation neural network on 881 PubChem substructures and reported the best fivefold-CV AUC of 0.8959 using drug-target features versus 0.7984 for SVM; crucially, AUPR stayed between 0.34 and 0.42 across all 16 method–feature combinations, so despite strong ranking, class imbalance in SIDER limits practical precision, and performance rests entirely on cross-validation within SIDER-derived data without external testing or code release. DSEP BP-neural-network ADR predictor

SpatialMOC reported PCC 0.8412 (RNA→ATAC) and 0.8221 (ATAC→RNA) on MISAR-seq E15.5 mouse embryo, 0.6859/0.6123 on SPOTS spleen RNA–ADT (1-FOSCTTM 0.8147/0.7845), ARI 0.8128 for mouse-brain spatial domains, and PCC 0.7646 under 20% masking — margins of +0.025 to +0.093 PCC over SpaMIE and scPair. Evaluation, however, used only public two-dimensional datasets with simulated dropout, and no code or model availability statement is provided. SpatialMOC cross-modality benchmark

Experimental studies: well-controlled but correlative

The ITP proteomics study quantified 613 of 829 proteins in bone-marrow mononuclear cells from 20 patients and 20 matched controls, and PRM confirmed HSPA8, PARK7, YWHAH and ITGB3 downregulation with CSF1R upregulation; its cross-sectional design, modest n, and lack of raw-data deposition leave causality untested. ITP bone-marrow autophagy proteomics

The pig crossbreeding study (100 animals; omics on six per breed) reported 616 DEGs and 681 metabolites, nominating FASN, SREBF1, ADIPOQ, IRS1 and ANGPTL4 as fat-metabolism hubs, with raw RNA-seq deposited under PRJNA1285098; integration is correlation-based, so gene–metabolite links (e.g., alphadimorphecolic acid with IRS1/ANGPTL4) await functional validation. Pig crossbreed fat-metabolism multiomics

The Chinese fir trial (three randomized blocks; nitrate vs ammonium; young vs two-year leaves; 562 metabolites) showed leaf-age-specific carbon–nitrogen reprogramming under nitrate, but a single site, a single season, and no explicit limitations section weaken generalizability. Chinese fir leaf-age nitrogen trial

The vitamin E trial (720 fish, six doses × four replicates, eight weeks) estimated a growth-based requirement of 100.37 mg/kg via Nrf2/Keap1 activation, while 198 mg/kg was detrimental; reliance on mRNA proxies and a single species limits transfer. Vitamin E aquaculture feeding trial

Cross-cutting critique

• Validation gap: both computational papers win modest margins on public, imbalanced or curated benchmarks without external or prospective testing.
• Reproducibility: DSEP and SpatialMOC provide no code; only the pig study deposited raw data.
• Causality: every experimental paper is observational or correlative at its integration step; no functional assays test the autophagy proteins or pig hub genes.
• What would change these conclusions: DSEP evaluated on post-2015 SIDER/FAERS additions; SpatialMOC on freshly generated paired tissue; knockdown experiments for HSPA8/CSF1R and FASN/SREBF1; multi-site replication for the fir trial.

Author Review: Shu Yuan Deep-dive: External ADR Validation

Run AI Scientist Analysis

Evolve Visual Paper Analysis

 BGPT Paper Review

Study Novelty

SpatialMOC's spatial-context-aware cross-modality factorization and DSEP's substructure-based neural network are incremental over prior multi-omics imputation and ADR-prediction work; the corpus's breadth is unusual but reflects evidence aggregation rather than a single conceptual breakthrough.

Scientific Quality

Reported benchmarks are internally consistent and one study deposited raw data, but both computational papers lack code and external validation, AUPR exposes severe class imbalance in DSEP, the ITP and pig studies are small-n and correlative, and the fir preprint lacks a limitations section.

Study Generality

SpatialMOC claims general cross-modality reconstruction across tissues, stages and platforms, and DSEP targets all drug classes, but each is demonstrated on a handful of public datasets while the experimental papers are species- or site-specific.

Study Usefulness

DSEP and SpatialMOC provide practical prediction tools with quantified benchmarks, the ITP proteins are candidate biomarkers, and the vitamin E requirement directly informs feed formulation; utility is capped by missing code and unvalidated candidate targets.

Study Reproducibility

Only the pig study deposited raw data (PRJNA1285098); DSEP and SpatialMOC provide no code or model artifacts, the ITP study deposited no raw spectra, and the fir trial's single-site design plus limited metabolomics replication hinders independent repetition.

Explanatory Depth

Mechanistic claims (mTOR/MAPK autophagy signaling, Nrf2/Keap1 activation, AMPK/PPAR remodeling) rest on pathway enrichment and correlation rather than perturbation experiments; the methods papers describe architectures but offer limited interpretability of learned representations.