Atopic Dermatitis Genetic Variant Study

Transcriptome-Wide Association Study (TWAS) of atopic dermatitis (AD) across 8 public GWAS cohorts (European, East Asian, Aboriginal Australian), using GTEx v8 mashr prediction models and the MetaXcan framework.

Datasets

Eight AD GWAS summary statistics sourced from OpenGWAS. See gwas-twas-analysis/manifest.csv.

OpenGWAS ID	Population	Year	Sample size
bbj-a-90	East Asian (BBJ)	2019	212,036
ebi-a-GCST90018564	East Asian	2021	168,103
ebi-a-GCST90018784	European	2021	481,299
ebi-a-GCST90027161	European	2021	796,661
finn-b-L12_ATOPIC	European (FinnGen)	2021	—
finn-b-ATOPIC_STRICT	European (FinnGen, strict definition)	2021	—
finn-b-ATOPIC_STRICT_REIMB	European (FinnGen, strict + reimbursement)	2021	—
ieu-b-5145	Aboriginal Australian (females)	2023	864,982

Pipeline (5 steps)

Raw GWAS (hg19 VCF)  ─[Step 1: harmonize]─>  GRCh38 tabular
                                                   │
                                         [Step 2: impute]
                                                   │
                                            220 batches
                                                   │
                                         [Step 3: merge]
                                                   │
                                      Imputed summary statistics
                                                   │
                                   [Step 4: S-PrediXcan × 49 tissues]
                                                   │
                                     49 tissue result files
                                                   │
                                   [Step 5: S-MultiXcan cross-tissue]
                                                   │
                                      Gene-level associations

Step 2 (imputation) is the expensive step — ~1,100 jobs total (8 traits × 22 chromosomes × 10 sub-batches = 1,760 max; we run the 5 previously-unfinished traits only on the cloud in practice). Deployed on GCP as 24 × n2-standard-8 VMs with a Filestore NFS job queue; see gwas-twas-analysis/scripts/twas/deploy/.

Results

Per-trait gene-level multi-tissue p-values are in gwas-twas-analysis/smultixcan/ — 8 CSVs × ~22,300 genes.

Per-trait single-tissue SPrediXcan results (49 tissues × 8 traits = 392 CSVs, ~720 MB total) are not checked in; regenerate locally by running Step 4.

Headline findings (bbj-a-90, East Asian)

34 genome-wide significant genes. Top hits cluster at:

• Chromosome 2 IL cluster: IL18RAP, IL18R1, IL1RL1 (established AD loci)
• Chromosome 6 MHC: HLA-DRB1, HLA-DQA1, HLA-DPB1, MICA, PSORS1C2, CCHCR1, POU5F1
• Complement: C2, C4B
• Others: RNF5 (p = 1.5×10⁻³¹ — top), GLB1, SLC9A4, TCF19

See notebooks/01_bbj_analysis.ipynb for the full walkthrough.

Repository layout

.
├── README.md
├── gwas-twas-analysis/
│   ├── manifest.csv                    — 8 datasets metadata
│   ├── smultixcan/                     — 8 × gene-level CSVs (~47 MB)
│   └── scripts/
│       ├── twas/                       — harmonize → impute → merge → SPrediXcan → SMulTiXcan
│       │   ├── config.sh, run_*.sh
│       │   ├── steps/                  — run_step{1..5}_*.sh
│       │   └── deploy/                 — GCP spin-up + autonomous launch + teardown
│       └── gwas/                       — OpenGWAS VCF → standardized table
└── notebooks/
    └── 01_bbj_analysis.ipynb           — full TWAS walkthrough for bbj-a-90

How to reproduce

# 1. Set up conda envs (metaxcan + gwasimputation)
bash gwas-twas-analysis/scripts/twas/setup_env.sh

# 2. Download OpenGWAS VCFs
bash gwas-twas-analysis/scripts/gwas/download_opengwas_ad.sh

# 3. Convert VCFs to standardized table format
bash gwas-twas-analysis/scripts/gwas/convert_vcf_batch.sh

# 4. Run the full TWAS pipeline
bash gwas-twas-analysis/scripts/twas/run_all_ad.sh

For cloud-autonomous imputation (Step 2) on GCP:

bash gwas-twas-analysis/scripts/twas/deploy/gcp_deploy_multi.sh <trait1> <trait2> ...

See gwas-twas-analysis/scripts/twas/deploy/ for the scale-up, async-launch, and download/teardown helpers.

Data availability

• GTEx v8 mashr prediction models: Zenodo 3518299
• 1000 Genomes LD reference panel: Zenodo 3657902
• OpenGWAS summary statistics: https://gwas.mrcieu.ac.uk

Citation

If you use this work, cite the underlying MetaXcan framework (Barbeira et al. 2018) and the individual GWAS studies listed in manifest.csv.