Corpus 1,272 assessed · 1,173 scored · 643 reproduced ≥75 · 168 flagged ·∅ 74.1/100
← New search

Cell type-specific eQTL analysis of COVID-19 based on single-cell transcriptomic data.

NAR Genom Bioinform · 2025
L1 51/100 3/4
⚑ Flagged for review — a reproduced result did not match the reported value

Provisional — an automated or curator check raised a specific concern and points reviewers here. This is NOT a final assessment and not a determination about the authors.

Why this verdict

The main results reproduced, with only marginal, non-material deviations.

Reproduced on the brainbox compute brainarbeit.com
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q5 · Derivability / plausibility 🟡
Content-critical question only partially held
+2 pts
From: Q7 · Core claim 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Input / endpoint not comparable 1:1
+1 pts
From: Q1 · Data identity 🔴
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +8
✓ What held up
  • Nothing in this column.
What did not (or only partly)
  • 🔴Could not use the authors’ exact input data
  • 🟡Reported values were only indirectly comparable
  • 🟡A deviation arose in the data or preprocessing
  • 🔴A deviation was attributed to the published material
  • 🟡Reported values were not (fully) derivable from the shared data
  • 🟡The deviation was non-trivial in magnitude
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
51/100
Reproducibility score
1.3 SD below mean
vs. all fields · 1173 studies
🎯 Scores higher than 13% of all assessed papers rank 1018 of 1173 scored

A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.

Reproduction agent’s raw note

Described well enough for the scRNA-seq front-end; not reproducible for the genotype-dependent core. Re-staged after «infra» reclaim via «our HPC» «job» (re-downloaded GSE165080 combined 10x matrix, Seurat 4.3.0.1). Dataset dimensions reproduce: 222457 cells (vs 222456, within-tol) and 53 samples (exact). The reported '22167 genes detected' is a MISMATCH: the deposited matrix detects 29232-39363 genes under standard min.cells thresholds (always MORE than 22167), so 22167 requires an unspecified extra gene filter - flagged for human review, not asserted as fabrication. Cell-type annotation (C4/C5) is only PARTIALLY reproducible: a standard canonical-marker labelling recovered 6 of the 8 reported broad immune types, but per-type counts diverge strongly (monocytes 72937 vs reported 24238; NK+Mono together 145k of 220k cells is biologically implausible for PBMC), because the authors' exact annotation procedure (markers/resolution/method) is not fully specified and a generic approach does not reproduce their counts. The eQTL/TWAS/Coloc/SMR core (C6/C7/C8) is OUT OF SCOPE and NOT attempted: repo+Zenodo ship R analysis code only; the SNP genotype matrix and per-cell-type expression those scripts consume are NOT deposited and no runnable variant-calling code is provided. Regenerating them needs raw FASTQ (SRP302338) through a large, under-specified GATK pipeline, beyond 80/20 scope.

These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.

Assessment versions

Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.

  1. v1 current initial assessment Score 56
    assessed: 2026-06-15 ⛓ 41a9bca64797
✎ I am an author of this paper

Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.

Reason for the rerun

We email you a confirmation link first. The rerun is an objective re-measurement — it cannot change the verdict in your favour, only ask us to look again.

Provenance — full disclosure

When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.

Reproduced
2026-06-23
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-16
no human curator yet
Last updated
2026-08-05

Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.

Deep full-text extraction

Model: sonnet
Founding hypothesis

The study tests whether common genetic variants exert cell type-specific and disease-stage-specific effects on gene expression in circulating immune cells during COVID-19, and whether such context-specific genetic regulation explains genetic susceptibility to COVID-19 identified by GWAS.

Core claims
  • Single-cell eQTL analysis across eight immune cell types identified 2593 genes whose expression is significantly associated with common genetic polymorphisms, with most genes showing cell type-specific effects finding
  • Monocytes show the most prominent genetic effects among immune cell types for COVID-19-risk variants identified by GWAS finding
  • 57 genes were identified as regulated by variants associated with COVID-19 via colocalization of eQTL and GWAS loci, with significant enrichment for dynamic effects across monocyte pseudotime/subtypes finding
  • Monocytes and platelets are markedly enriched in COVID-19 patients compared to healthy controls, with monocyte abundance increasing with disease severity finding
  • MIF and PARs signaling pathways are COVID-19-exclusive intercellular communication features, while APRIL and EPHB signaling are uniquely maintained in healthy controls; MIF signaling is dominated by monocyte-directed production from T cells via MIF-CD74/CD44 finding
  • Colocalizing genes involved in the protein interaction network are enriched in interferon-mediated signaling pathways mechanism
  • Bulk tissue eQTL resources (e.g. GTEx) mask cell type-specific regulatory variants due to averaging of expression across cell types within tissue finding
  • A pipeline for calling genetic variants directly from scRNA-seq reads (fastp/STAR/GATK) combined with matrixEQTL-based cis-eQTL mapping enables cell type-specific and stage/pseudotime-specific eQTL discovery method
Experimental setups
Assay System Perturbation Readout Platform
scRNA-seq (10x/Cell Ranger) PBMCs from 42 COVID-19 cases (5 asymptomatic, 13 mild, 12 severe, 12 convalescent) and 11 healthy controls COVID-19 infection/disease stage cell type composition, gene expression, marker genes Cell Ranger v5.0.0; Seurat v4.0.5; GRCh38/Ensembl v43
Genetic variant calling from scRNA-seq reads PBMCs, same 53 samples none high-quality autosomal SNPs fastp v0.23.2, STAR v2.7.10, GATK v4.2.6.1, snpEff v5.1d
cis-eQTL mapping Eight immune cell types (CD4+ T, CD8+ T, B cells, monocytes, cDC, pDC, platelets) from PBMCs genotype (SNP) vs gene expression, adjusted for age/sex eQTL associations (genotype-expression), significant eGenes matrixEQTL v2.3.0
Cell-cell communication analysis PBMC immune cell types, COVID-19 patients vs healthy controls COVID-19 infection ligand-receptor communication probability across 36 signaling pathways CellChat v1.6.1 with CellChatDB
Pseudotime trajectory inference 24238 monocytes (15000 classical, 9283 nonclassical) none (differentiation state) pseudotime ordering, stage/pseudotime-dependent dynamic eQTLs monocle v2.22.0
GWAS-eQTL colocalization eQTL data per cell type combined with COVID-19 GWAS (18152 cases, 1145546 controls) none colocalization probability between eQTL and GWAS loci coloc v5.1.0
Transcriptome-wide association study (TWAS) Cell type-specific transcriptome reference panels with 1000 Genomes LD reference none imputed genetic expression, novel gene-COVID-19 associations FUSION software
Gene ontology/pathway enrichment analysis Top 20 marker genes per cell type from PBMC scRNA-seq none enriched GO terms/pathways clusterProfiler v4.0.5
Key results
  • 222 456 high-quality cells and 22 167 genes passed QC, derived into eight major immune cell types
  • 4549 eQTLs identified for 2593 autosomal genes across eight cell types; 4284 were cell type-specific eQTLs 4549 eQTLs / 2593 genes
  • Per cell type, 356-1509 significant eQTLs detected, of which 281-1026 were unique to that cell type 356-1509 eQTLs
  • Replication against GTEx bulk tissue data was moderate, with high allelic direction consistency 6.6%-43.2% eQTL replication, 2.4%-23.3% eGene overlap, 94.1%-100% direction consistency
  • Replication against DICE single-cell eQTL database across major immune cell types 16.1%-28.1% replication, 96.3%-99.1% direction consistency
  • CD4+ T cells showed highest replication rate against an independent PBMC sc-eQTL dataset with perfect direction consistency 20.8% replication, 100% direction consistency
  • pDCs had minimal eQTL overlap with an independent PBMC sc-eQTL dataset but strong effect size correlation n=9 overlapping eQTLs, Pearson r=0.89, P<.001
  • 153 527 high-quality autosomal SNPs obtained after variant calling and filtering pipeline 153 527 SNPs
Key statistics
  • count 222 456 cells, 22 167 genes (cells/genes passing scRNA-seq QC)
  • count 4549 eQTLs for 2593 genes (total eQTLs identified across 8 cell types)
  • count 153 527 autosomal SNPs (high-quality SNPs after variant calling pipeline)
  • other 6.6%-43.2% eQTL replication, 2.4%-23.3% eGene overlap (replication vs GTEx bulk data across 8 cell types)
  • other 16.1%-28.1% replication (replication vs DICE single-cell eQTL database)
  • other 20.8% replication, 100% direction consistency (CD4+ T cells vs independent PBMC sc-eQTL dataset)
  • correlation Pearson's r = 0.89, P < .001 (pDC eQTL effect size correlation with independent PBMC sc-eQTL dataset)
  • count 18 152 COVID-19 samples and 1 145 546 controls (COVID-19 Host Genetics Initiative GWAS summary data used for colocalization)

Statistical methods review

Model: sonnet

A neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.

The paper profiled >222,000 PBMCs from 42 COVID-19 cases (four severity stages) and 11 healthy controls using scRNA-seq, called SNPs directly from the RNA-seq reads via GATK, and performed cis-eQTL mapping in eight immune cell types with a linear model (matrixEQTL) adjusting for age and sex within a 1 Mb window around each gene's TSS. Statistical significance was set at a nominal P < 0.05 with Benjamini-Hochberg FDR correction (FDR < 0.05) for the eQTL analyses. Identified eQTLs were integrated with COVID-19 GWAS summary statistics via Bayesian colocalization (coloc) and TWAS (FUSION) to nominate causal genes, and pseudotime-dependent eQTLs in monocytes were explored by dividing the trajectory into nine bins.

Replicationbiological Sample size53 individuals total stated; 42 COVID-19 (5 asymptomatic, 13 mild, 12 severe, 12 convalescent) and 11 healthy controls; no formal power calculation mentioned GroupsCOVID-19 cases (four severity stages) vs. healthy controls; across eight immune cell types Pairingunpaired Randomization/blindingnot stated Dispersionnone Exact p-valuesno Effect sizesyes Confidence intervalsno Multiplicity correctionBenjamini-Hochberg FDR
Statistical tests used
Test Applied to n Assumptions
Linear regression (matrixEQTL) Cis-eQTL mapping in each of eight immune cell types, testing SNP-gene associations within 1 Mb of each gene's TSS 53 individuals (42 COVID-19, 11 healthy); 153,527 autosomal SNPs; 22,167 genes not stated
Student's t-test Differential gene expression for selection of top 20 marker genes per cell type for enrichment analysis 222,456 cells across 8 cell types not stated
Bayesian colocalization (coloc v5.1.0) Integration of cis-eQTL signals with COVID-19 GWAS loci within 500 kb windows around lead eQTL variants GWAS: 18,152 COVID-19 cases, 1,145,546 controls; eQTL n=53 not stated
TWAS (FUSION) Imputation of cell type-specific genetic expression to identify novel COVID-19 gene associations null not stated
Pearson's r correlation Effect size correlation for replicated eQTLs in pDC validation against external dataset n=9 overlapping eQTLs in pDC comparison not stated
Pseudotime trajectory inference (Monocle v2.22.0, differentialGeneTest function) Identification of highly variable genes driving monocyte pseudotime ordering; 24,238 monocytes 24,238 monocytes (15,000 classical, 9,283 nonclassical) na
Approaches that could also have been used
  • Genotypes were derived from scRNA-seq reads using a GATK RNA-seq variant-calling pipeline
    Could also: Matched whole-genome or whole-exome sequencing from the same donors could also be used for genotyping — Genotyping from genomic DNA avoids allele-specific expression bias, uneven transcriptomic coverage across the genome, and the confounding of somatic/technical variants with germline SNPs; it is the conventional approach in population-level eQTL studies and allows denser genome-wide variant sets
  • Cis-eQTL mapping used pseudobulk mean expression per gene per individual per cell type as input to a linear model
    Could also: Mixed models explicitly accounting for within-individual cell-level correlation (e.g., SAIGE-QTL, tensorQTL with random effects, or negative-binomial mixed models) could also be used — Single-cell-aware approaches model the overdispersion of count data and the varying number of cells per donor, which can improve power when donor cell counts are unbalanced and reduce inflation of test statistics compared to treating pseudobulk means as Gaussian observations
  • Context-specific eQTLs were identified by subsetting monocytes by disease stage or pseudotime bin and re-running the linear model independently in each subset
    Could also: A genotype-by-context interaction term (genotype × stage or genotype × pseudotime) in a single unified model could also formally test for differential eQTL effects — An interaction model directly tests whether eQTL effect sizes differ across conditions in a single framework, avoids the reduction in effective sample size from subsetting, and provides a more powerful and calibrated test of context-specificity
  • eQTL significance was assessed with a nominal P-value threshold followed by Benjamini-Hochberg FDR across all tested SNP–gene pairs
    Could also: Permutation-based FDR (e.g., as in FastQTL or tensorQTL) computing empirical per-gene P-values before cross-gene FDR correction could also be applied — Permutation approaches account for the variable number of SNPs tested per gene (which makes the null distribution of minimum p-values gene-dependent) and can improve FDR calibration, particularly when cis-windows contain highly correlated SNPs
  • Marker gene selection used Student's t-test on single-cell expression values
    Could also: Wilcoxon rank-sum test or pseudobulk methods (e.g., DESeq2 or edgeR on donor-aggregated counts) could also be used for scRNA-seq differential expression — The Wilcoxon rank-sum test makes fewer distributional assumptions about count data; pseudobulk methods additionally account for biological variability between donors rather than treating individual cells as independent observations
  • Replication across external eQTL datasets was quantified by overlap counts and proportion of shared variants with consistent effect direction
    Could also: The π1 statistic or regression of discovery effect sizes against replication effect sizes across all tested pairs (not just significant ones) could also quantify replication — These approaches use information from all tested SNP–gene pairs and are less sensitive to arbitrary significance thresholds applied in the replication dataset, providing a more continuous measure of overall replication signal
Software: matrixEQTL 2.3.0 · coloc 5.1.0 · FUSION (TWAS) · Seurat 4.0.5 · Cell Ranger 5.0.0 · clusterProfiler 4.0.5 · CellChat 1.6.1 · Monocle 2.22.0 · GATK 4.2.6.1 · STAR 2.7.10 · fastp 0.23.2 · snpEff 5.1d · R 3.6.1

Citation network

Where this publication sits in the reproducibility-weighted citation graph — what it is built on, and what is built on it. Citation data from OpenAlex.

Citations
0
Impact: low
Foundation confidence
None of its references are in our reproducibility record yet — its foundation cannot be assessed.
Topics

No assessed neighbours yet — the network grows as more papers are assessed.

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-41064594

Cell type-specific eQTL analysis of COVID-19 based on single-cell transcriptomic data (Wang et al., NAR Genom Bioinform 2025)

Pipeline map (per reported result)

Result Pipeline Inputs shipped? In scope?
Post-QC cell/gene counts (222,456 cells, 22,167 genes); 53 samples Seurat QC on GSE165080 combined matrix YES (GEO combined matrix.mtx.gz + barcodes/genes/cell_batch) YES (primary)
8 major immune cell types + proportions (e.g. 24,238 monocytes) Seurat cluster + canonical-marker annotation (script 1) YES (same matrix) YES (attempt)
4549 cis-eQTLs / 2593 genes; 356–1509 per cell type Matrix eQTL (script 6) NO — needs SNP genotype matrix (153,527 SNPs) called from RNA-seq via GATK Best-Practices + per-cell-type pseudobulk expression; neither shipped, no runnable variant-calling code NO (blocked)
TWAS 43 genes (script 7, FUSION) FUSION + eQTL weights NO — depends on eQTL sumstats above NO (blocked)
Coloc 19 loci (8.COLOC.md), SMR (9.SMR.md) coloc/SMR vs HGI GWAS NO — depends on eQTL sumstats above NO (blocked)
CellChat networks (script 3), Monocle monocyte trajectory (script 5) viz/trajectory on annotated object partial (needs annotated object) NO (low priority, not pipeline-derived numbers)

Rationale

The repo + Zenodo ship analysis code only. The genotype-dependent core (eQTL/TWAS/Coloc/SMR) consumes intermediate files (genotype.txt = 153,527-SNP matrix, per-cell-type expression) that are NOT deposited and whose generation (RNA-seq variant calling via GATK) is not provided as runnable code — only described in Methods. Regenerating them requires raw FASTQ (SRP302338) through a large, under-specified GATK pipeline. Out of scope per 80/20 + docs-insufficient for those inputs. The scRNA-seq front-end (QC counts + cell-type annotation) IS reproducible because the combined post-QC matrix is on GEO and script 1 consumes exactly that 10X-format input.

Figures / tables: Fig1Table
C1
Reported
222456 high-quality cells
Reproduced
222457
within tolerance
C2
Reported
53 PBMC samples (42 COVID + 11 HC)
Reproduced
53
exact
C3
Reported
22167 genes detected
Reproduced
39363 (>=1c) / 34335 (>=3c) / 29232 (>=10c)
did not match
C4
Reported
8 major immune cell types
Reproduced
6 broad types (B, CD4 T, CD8 T, Monocyte, NK, Megakaryocyte)
partial
C5
Reported
24238 monocytes
Reproduced
72937
did not match
C6
Reported
4549 cis-eQTLs / 2593 genes
Reproduced
not attempted (out of scope)
partial
C7
Reported
43 TWAS genes
Reproduced
not attempted (out of scope)
partial
C8
Reported
19 colocalizations
Reproduced
not attempted (out of scope)
partial

Assessments & scoring basis

Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.

🤖 AI curator · claude (ai-curator room) · v1.0 L1 51/100

An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.

🔴1. Data identity
🟡2. Endpoint comparability
🟡3. Location of the main deviation
🔴4. Cause of the deviation
🟡5. Derivability / plausibility
🟡6. Severity of the deviation
🟡7. Core claim
🟡8. Severity of the miss (overall human judgment)
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q5 · Derivability / plausibility 🟡
Content-critical question only partially held
+2 pts
From: Q7 · Core claim 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Input / endpoint not comparable 1:1
+1 pts
From: Q1 · Data identity 🔴
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +8

The scRNA-seq front-end reproduces well: 222,457 cells (vs 222,456, off-by-one) and 53 samples (exact) come straight from the deposited GSE165080 matrix. However C3 '22,167 genes detected' is not recoverable — the public matrix yields 29,232–39,363 genes under all standard thresholds, implying an undocumented gene filter (authors-side method gap, flagged but not fabrication). The headline eQTL/TWAS/coloc core (C6–C8) is entirely non-reproducible because the RNA-seq-derived genotype matrix and its variant-calling code were never deposited — a data-availability/code-incompleteness issue, not a demonstrated contradiction. Overall: a faithful but partial reproduction with explainable deviations; the central conclusion remains unverified rather than refuted.

🤝
Reproduced automatically — and fairly

Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.

Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.

🚩 Report an error in this record

Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.

Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.

Reproduction footprint

claude-opus-4-8

Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.

236.8 k
tokens (I/O) · 13.2 M incl. cache
78 min
runtime · 0.23 CPU-h
26.2 GB
peak RAM
1
HPC jobs
hummel
machine