Check Claims Before You Publish

 Quick Explanation Copied

 Long Explanation

Author Identity and Citation Metrics

The queried Semantic Scholar record (h-index 0, 2 papers, including a 19th-century pharmacology paper "Zur Pharmakologie der Sabina") clearly reflects a name-collision artifact and does not describe the structural-biology author of the three 2026 preprints. OpenAlex disambiguates two "Yuki Kagaya" profiles; the relevant one (ORCID 0000-0003-0146-1709) has 42 works, 1,021 citations, h-index 13, with a publication surge in 2025–2026 (11 + 9 works) and topic scores peaking in Biology, AI, and protein structure prediction. The most-cited works are coexpression database papers (ATTED-II 2018: 252 citations; COXPRESdb v7: 138 citations) authored at Tohoku University, while recent first-author work (NuFold, Nature Communications, 66 citations) is at Purdue University with Daisuke Kihara NuFold RNA structure prediction paper. The h-index of 13 indicates a mid-career, productive contributor rather than a field leader — respectable for a database/methods developer, modest relative to senior computational biologists.

Caveat: 2026 citations (5) are incomplete by definition; the 2017–2018 peaks are database-paper citations accrued over time, not recent activity.

Scientific Strength of the 2026 Preprint Portfolio

Kagaya is a co-author on three 2026 preprints from the Kihara lab, each a substantial computational method:

• CryoZeta (Feb 2026): diffusion-based cryo-EM model building achieving complex TM-score 0.933 vs AF3's 0.794, protein 0.973 vs 0.871, DNA/RNA 0.678 vs 0.383 on 200 test maps; protein sequence recall 0.957 vs 0.83 for other methods. Strongly benchmarked with careful non-redundant map-group splits (1,106 train / 138 val / 139 test groups) and a public web server CryoZeta benchmark on 200 cryo-EM maps.
• Prot-LAMBDA (Aug 2026): distance-aware PLM roughly doubling long-range contact recall (CASP14 0.39 vs 0.17–0.19; CAMEO 0.60 vs 0.24–0.27); yet the downstream LambdaFold pipeline underperforms ESMFold overall (TM 0.71 vs 0.78), honestly reported by the authors — a positive sign of rigor LambdaFold vs ESMFold benchmark results.
• PathFold (Aug 2026): sequence-only folding-pathway prediction with PathScorer alignment 0.804–0.815 vs SMD (reference SMD-SMD self-agreement 0.894). Weakest of the three — validation is against SMD simulations, not experimental folding pathways; only two proteins checked against Φ-values; reproducibility scored lowest (4/10) PathFold folding pathway validation.

Blindspots and Confidence

Role clarity is a limitation: in the three 2026 preprints Kagaya's precise authorship position is not specified in the supplied data, so credit cannot be attributed individually. All three works share NIH/NSF funding to D. Kihara; CryoZeta declares an Intellicule, LLC founding-membership conflict for two co-authors CryoZeta funding and COI statement. The evidence supports a picture of a technically strong, method-generating computational biologist whose strongest work is in cryo-EM modeling and databases, with folding-pathway work still needing experimental validation.

 BGPT Author Review

Scientific Quality

H-index 13 with 1,021 citations and 42 works places Kagaya as a solid mid-career contributor. The flagship ATTED-II/COXPRESdb database papers (252, 138 citations) show sustained community impact, and NuFold (first-author, Nature Communications) demonstrates independent method leadership in RNA structure prediction. The 2026 preprints are technically ambitious (diffusion cryo-EM modeling, distance-aware PLMs, folding pathways) with strong benchmarking in CryoZeta. Weaknesses: PathFold validation is simulation-only with weak reproducibility; LambdaFold underperforms its main baseline overall; Kagaya's exact contribution role in the 2026 works is unverifiable from supplied data; and the Semantic Scholar profile fragment (h=0) suggests past name-disambiguation problems inflating apparent early-career confusion.

Communication Quality

The supplied papers report quantitative results clearly (TM-scores, hit rates, contact recall with explicit comparators), provide web servers and code repositories (em.kiharalab.org, GitHub, Zenodo), and honestly report negative results such as LambdaFold trailing ESMFold and YidC early-folding mismatch. This transparency is commendable. Deductions: heavy jargon density typical of the field, and PathFold lacks a data availability statement, limiting reader follow-up.

Author Novelty

CryoZeta's diffusion-based integration of cryo-EM density with AF3-style sequence prediction for protein-nucleic acid complexes is a genuinely new capability (RNA TM-score 0.678 vs AF3's 0.383). PathFold's sequence-only folding-pathway prediction is a novel problem framing. Prot-LAMBDA's explicit distance supervision in PLM pretraining is an interesting but incremental idea in a crowded field. Overall novelty is high, driven by first-of-kind applications rather than fundamentally new theory.

Scientific Rigor

CryoZeta shows the strongest rigor: 200 test maps, non-redundant group-level splits, multiple baselines, ablation variants, resolution-range analysis. Prot-LAMBDA honestly reports underperformance on the full benchmark. PathFold is the weak point: circular validation against the same SMD simulation paradigm used for training data, no experimental kinetics validation beyond two proteins' Φ-values, low reproducibility score, and no explicit limitations section in the supplied record. Potential overfitting to curated PDB/EMDB training distributions is acknowledged but not fully quantified.