Check Claims Before You Publish

 Quick Explanation Copied

 Long Explanation

Career Profile and Citation Impact

Yao-Qing Chen (ORCID 0000-0002-4659-1550) presents a well-established publication record: 117 works, ~3,391 citations, and h-index 28 per OpenAlex (vs h-index 13/736 citations in a narrower Semantic Scholar profile, illustrating common author-disambiguation fragmentation). Citation velocity peaked in 2018–2021, driven by first-author work in Cell on influenza neuraminidase-cross-reactive antibodies (440 citations) and contributions to Cell Host & Microbe and Science Translational Medicine. Top work spans influenza B-cell immunology, MERS-CoV structural biology (first-author crystal structure of the MERS-CoV RBD), SARS-CoV-2 immunology cohorts, and environmental disinfection chemistry — a breadth that is both a strength (cross-disciplinary fluency) and a potential blindspot (no single deep specialty in machine learning, relevant to the computational AMPBAN paper).

Scientific Strength of the AMPBAN 2026 Preprint

The most recent work under review, AMPBAN, applies ESM3 sequence embeddings fused with ESMFold-predicted structures via an EGNN and Bilinear Attention Network for antimicrobial peptide prediction. Strengths: rigorous leakage removal (103,561 initial AMP entries curated down to 30,971 AMP / 38,675 non-AMP training sequences), three balanced independent test sets (XIAO 920/920, MFA 615/606, XU 1,536/1,536), full metric reporting (ACC, MCC, Sn, Sp, F1, AUC with 5-fold CV), open code/data on GitHub, and declared no competing interests AMPBAN Benchmark and Ablation Results.

Critical Assessment and Blindspots

Reported observation: generalization drops sharply on independent data — MCC falls from 0.8606 to 0.6887 and sensitivity from 0.9463 to 0.7258, with the XU set ablation at ACC 0.7077/MCC 0.4691 indicating dataset-specific overfitting the authors themselves acknowledge AMPBAN Declared Limitations. What is missing: zero wet-lab validation of predicted AMPs, purely computational evidence, and reliance on predicted (not experimental) structures may bias atypical peptides. The work is a preprint (not peer-reviewed as of the given record), dated January 2026. Inference (BGPT): Chen is a competent collaborator bridging experimental virology with computational biology, but this paper's ML novelty rests on established components (ESM3, ESMFold, EGNN, BAN); the decisive test — prospective experimental AMP activity — remains undone, and falsification would require demonstrating failure on truly novel peptide families or equivalence of structure-free variants.

 BGPT Author Review

Scientific Quality

Established mid-career researcher with h-index 28 and ~3,391 citations, including a first-author Cell paper on influenza NA antibodies (440 citations) and first-author structural work on MERS-CoV RBD. The AMPBAN preprint shows solid data curation and honest metric reporting but is computational-only, builds on existing components, and has no experimental validation. Breadth across virology, immunology, and environmental chemistry suggests versatility but dilutes deep methodological ML expertise.

Communication Quality

The AMPBAN paper reports complete metrics, clear architecture descriptions, and an accessible GitHub repository. Openly declared limitations suggest transparency. However, the record does not permit assessment of prose clarity directly, and dense multi-modal fusion descriptions may challenge non-specialists.

Author Novelty

Incremental methodological novelty in AMPBAN but historically high novelty in antibody/virus structural work.

Scientific Rigor

Strong leakage-removal protocol (103,561 curated to ~69,600 training sequences), three independent balanced test sets, 5-fold CV, full metric panels, open code, and self-acknowledged overfitting on some datasets all indicate rigor. Points deducted for absence of experimental validation and reliance on predicted structures.