Cancer Biology · Nature Communications 2024

AI Diagnostic vs. Four Expert Reviewers

A 55-page paper with single-cell RNA-seq, ATAC-seq, and zebrafish xenotransplantation data. Nature Communications assigned four reviewers. We ran our diagnostic blind, then compared every finding.

5/5

Major issues matched

+1

Issue reviewers missed

19 min

Analysis time

$49

Cost

3,054

Words of feedback

The paper

“A human neural crest model reveals the developmental impact of neuroblastoma-associated chromosomal aberrations”

Saldana-Guerrero et al. present an hESC-based model for studying how chromosomal copy number alterations and MYCN overexpression drive neuroblastoma tumor initiation during neural crest development. Four isogenic cell lines, five differentiation timepoints, single-cell RNA-seq of 45,949 cells, ATAC-seq across 51 samples, and zebrafish xenotransplantation.

Issue-by-issue comparison

Missing MYCN-only control line

Matched

Manusights AI Diagnostic

Listed as Essential Experiment A: can't attribute phenotype to CNV-MYCN cooperation without a WT+MYCN comparator.

Nature Communications Reviewers

All 4 reviewers flagged this. R1: "should be included." R3: "critical comparators." R4: "An important reference is lacking."

Cannot separate CNV vs. MYCN contributions

Matched

Manusights AI Diagnostic

Identified the mechanistic attribution gap: NR1D1 and TFAP4 are known MYCN targets, so chromatin data partially recapitulates known biology.

Nature Communications Reviewers

Reviewers 1, 3, and 4 all raised this. R4: "might be due to MYCN overexpression alone."

Single hESC line limits generalizability

Matched

Manusights AI Diagnostic

Flagged single H7 line as a weakness. Recommended second hESC line as a prioritized experiment.

Nature Communications Reviewers

R3 raised sex-related incidence differences and generalizability concerns for a single female line.

Weak xenograft controls

Matched

Manusights AI Diagnostic

Identified missing controls in zebrafish xenotransplantation experiments.

Nature Communications Reviewers

R1: "weak point." R2: "show positive controls." R3: "were tumors transplantable?"

Missing n numbers and error bars

Matched

Manusights AI Diagnostic

Flagged incomplete statistical reporting across multiple figure panels.

Nature Communications Reviewers

R2 requested explicit quantification. R1 asked for detailed genomic analysis confirmation.

Overlap with competing Fang et al. (2025) publication

AI only

Manusights AI Diagnostic

Identified significant overlap with Fang et al. (2025), who independently built an hESC-based 17q model. Recommended cross-referencing their six candidate driver genes against the dataset.

Nature Communications Reviewers

None of the four reviewers flagged this. Fang et al. was published after the review took place.

What reviewers caught that we missed

Four items. All presentation-level or minor methodological. None would independently cause rejection.

Doxycycline-only control (reagent-specific artifact check)
CNV vs. CNA terminology consistency
Supplementary video completeness (2 of 4 genotypes filmed)
Motility claim lacked quantification
“The AI diagnostic matched every major scientific concern raised by four expert reviewers, and identified a competing publication none of them caught.”

19 minutes. $49. Blind analysis.

What the diagnostic report contained

7-dimension quality scorecard

Originality, claims, soundness, clarity, value, prior work, importance

7 specific weaknesses identified

MYCN attribution, single hESC line, Jaccard limitations, missing n, incomplete comparisons, batch effects, heatmap units

5 specific strengths

Isogenic design, multi-omic integration, zebrafish validation, tumor mapping, TF network analysis

6 recommended experiments

2 essential (MYCN-only line, TF loss-of-function) + 4 recommended, each with specific protocols

Journal fit assessment

Nature Communications appropriate; specified bar for Nature Medicine or Nature Genetics

7 verified citations

Including Fang et al. (2025), the competing publication no reviewer flagged