henhaobio
Sign in

Cell GenomicsGenetic & Genomic MedicinePeer-reviewed journal article

Global Biobank Meta-analysis Initiative: powering genetic discovery across human diseases

Wei ZhouiD, Masahiro KanaiiD, Kuan-Han WuiD, Humaira RasheediD, Kristin TsuoiD, Jibril B HirboiD, Ying WangiD, Arjun BhattacharyaiD, Huiling ZhaoiD, Shinichi NambaiD, Ida L. SurakkaiD, Brooke N. WolfordiD, Valeria Lo FaroiD, Esteban A. Lopera-MayaiD, Kristi LälliD, Marie-Julie FavéiD, Juulia PartaneniD, Sinéad B. ChapmaniD, Juha M KarjalaineniD, Mitja KurkiiD, Mutaamba Maasha, Ben Michael BrumptoniD, Sameer ChavaniD, Tzu‐Ting CheniD, Michelle DayaiD, , Jennifer E. HuffmaniD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Biobanks are being established across the world to understand the genetic, environmental, and epidemiological basis of human diseases with the goal of better prevention and treatments. Genome-wide association studies (GWAS) have been very successful at mapping genomic loci for a wide range of human diseases and traits, but in general, lack appropriate representation of diverse ancestries - with most biobanks and preceding GWAS studies composed of individuals of European ancestries. Here, we introduce the Global Biobank Meta-analysis Initiative (GBMI) -- a collaborative network of 19 biobanks from 4 continents representing more than 2.1 million consented individuals with genetic data linked to electronic health records. GBMI meta-analyzes summary statistics from GWAS generated using harmonized genotypes and phenotypes from member biobanks. GBMI brings together results from GWAS analysis across 6 main ancestry groups: approximately 33,000 of African ancestry either from Africa or from admixed-ancestry diaspora (AFR), 18,000 admixed American (AMR), 31,000 Central and South Asian (CSA), 341,000 East Asian (EAS), 1.4 million European (EUR), and 1,600 Middle Eastern (MID) individuals. In this flagship project, we generated GWASs from across 14 exemplar diseases and endpoints, including both common and less prevalent diseases that were previously understudied. Using the genetic association results, we validate that GWASs conducted in biobanks worldwide can be successfully integrated despite heterogeneity in case definitions, recruitment strategies, and baseline characteristics between biobanks. We demonstrate the value of this collaborative effort to improve GWAS power for diseases, increase representation, benefit understudied diseases, and improve risk prediction while also enabling the nomination of disease genes and drug candidates by incorporating gene and protein expression data and providing insight into the underlying biology of the studied traits.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (21 Nov 2021) · CC BY 4.0

Meta-analysis of harmonized GWAS from up to 18 biobanks (six ancestry groups) across 14 endpoints replicated 320 known loci and identified 188 apparently novel loci.

Most GWAS and biobanks are dominated by European-ancestry participants, and less prevalent or electronic health record (EHR)-derived endpoints such as procedures have rarely been studied genetically. GBMI shows that biobanks differing in recruitment, case definitions, genotyping, and ancestry can be pooled via harmonized phecode-based phenotypes and a shared GWAS and QC pipeline, raising discovery power, ancestral diversity, and polygenic risk prediction.

Key results

  1. ShownPooling biobanks yielded 508 genome-wide loci across the 14 endpoints, and nearly all lead variants showed homogeneous effects across biobank and ancestry data sets despite differing recruitment and phenotyping.

    Inverse-variance meta-analysis: 320 known plus 188 novel loci from 70.8 million tested variants; ~96% of 508 lead variants showed no heterogeneity (Cochran's Q p ≥ 1/508); 87 loci had a protein-coding lead or LD proxy (r^2 > 0.8), 18 novel; MR-MEGA added 17 loci across 10 endpoints (12 novel); liability-scale SNP heritability 1.79% to 10.73% (13 of 14 endpoints). · All-Biobank Meta-analysis

  2. ShownNon-European samples widened the variant space and yielded loci with index variants far more common outside Europe, e.g., East Asian-enriched signals for primary open-angle glaucoma (POAG), COPD and asthma and African-enriched ones for venous thromboembolism (VTE), heart failure and stroke.

    European-only meta-analysis: 343 loci; adding non-European samples gave 165 more. Index variants ≥10-fold more frequent in non-European ancestries (EUR frequency < 5%): 21 of 165 (12.7%) vs 4 of 343 (1.16%). 21.8 million extra variants tested (18 million with MAF ≤ 1%); nine of 499 multi-ancestry loci were heterogeneous across ancestry. · Power improved by incorporating samples with…

  3. ShownLeave-one-biobank-out (LOBO) comparisons and replication of earlier consortia support integrating heterogeneous biobanks, with deviations concentrated in smaller, multi-ancestry or South Asian-including biobanks and in POAG defined by self-report.

    Biobank vs LOBO Deming slopes (index variants p < 1×10^−10) mostly did not differ from one; genetic correlations with LOBO were near 1 for asthma, gout and COPD. Concordant effect directions: all 18 Trans-National Asthma Genetic Consortium (TAGC) asthma loci, 24 of 25 MVP AAA loci, 40 of 40 prior gout loci (three GBMI biobanks, about 20% of GBMI samples, overlapped the prior gout study). · Integration of association results across…

  4. ShownLarger case counts boosted discovery for common (asthma) and understudied (gout) diseases, pooling with existing consortia added POAG loci, and an EHR procedure endpoint (appendectomy) recovered acute appendicitis signals.

    Asthma: 153,763 cases and 1,647,022 controls vs TAGC's 23,948 cases and 118,538 controls; 180 loci (49 novel). Gout across 15 biobanks: 37,105 cases vs 13,179 in the largest prior study; 52 loci (30 novel), 40 linked to serum urate. POAG with two prior meta-analyses: 103 loci (19 novel). Acute appendicitis: 3 of its 10 loci also significant for appendectomy. · Improving power of genetic discovery for common…; Providing opportunities for genetic studies on…

  5. ShownFour gene-prioritization approaches (DEPICT, PoPS, TWAS, proteome-wide MR) each nominated plausible disease genes but overlapped little, so consensus candidates were rare.

    Asthma: 618 genes from at least one method, none from all four, 5 from three (FCER1G, IL18R1, IL4R, SMAD3, IL2RB). POAG: 204 genes, none from three or more. VTE: 244 genes; F2 supported by all four, and 5 genes (F5, PLCG2, PLEK, PROC, PROS1) by three. · Prioritization of cell types, tissues, and genes

Methods, data and reproducibility

Biobanks ran ancestry- and sex-stratified GWAS (SAIGE or REGENIE recommended) on phecode/OPCS-harmonized endpoints; a central team performed gnomAD-based QC and fixed-effect inverse-variance and MR-MEGA meta-analyses, then leave-one-biobank-out checks, LDSC, phenome-wide look-ups, gene prioritization and PRS-CS scoring.

  • GBMI consortium of 19 EHR-linked biobanks on four continents (e.g., UKBB, FinnGen, BBJ, CKB, BioVU, HUNT, MGI), mixing population- and hospital-based recruitment · human · n = more than 2.1 million individuals
  • Participants in six ancestry groups: EUR (1.4 million), EAS (341,000), AFR (~33,000), CSA (31,000), AMR (18,000), MID (1,600) · human · n = 6 main ancestry groups
  • Ancestry- and sex-stratified GWAS summary statistics for 14 endpoints, e.g., asthma, COPD, HF, stroke, gout, VTE, POAG, AAA, IPF, thyroid and uterine cancer, appendectomy · human · n = up to 18 biobanks; 70.8 million variants tested
  • 1000 Genomes and HGDP reference samples used to derive PC weights for projecting all biobanks into one ancestry space · human · n = 168,899 overlapping variants
  • GTEx v8 genotypes and expression (lung, brain cortex, aortic/coronary/tibial artery, heart) for training TWAS expression models · human · n = 296 European donors
  • ARIC European-ancestry pQTL instruments for proteome-wide Mendelian randomization on eight GBMI diseases · human · n = 5,418 pQTLs for 1,310 proteins
  • UK Biobank phecode GWAS used for phenome-wide look-ups of GBMI index variants · human (UK Biobank) · n = 1,283 diseases (1,238 in Results)
  • Prior comparator GWAS: TAGC asthma meta-analysis (66 cohorts), MVP AAA loci, largest prior gout meta-analysis (20 studies) · human · n = TAGC: 23,948 cases, 118,538 controls; gout: 13,179 cases
Methods & tools
SAIGE / REGENIE (recommended for per-biobank GWAS), Fixed-effect inverse-variance meta-analysis; Cochran's Q, MR-MEGA, Deming regression (biobank vs leave-one-biobank-out), LDSC; Popcorn, DEPICT, PoPS with MAGMA, JTI, MOSTWAS and FOCUS (TWAS), Proteome-wide MR (gIVW, MR-Egger, PWCoCo, MR-Steiger), PRS-CS; pruning and thresholding (P+T), ANNOVAR; VEP with LOFTEE (Hail), gnomAD (allele-frequency QC reference)
Software & environment
Meta-analysis on GRCh38 (GRCh37 for UKBB look-ups); GTEx ver. 8; 1000 Genomes Phase 3 European LD for PRS-CS; no software versions or containers stated.
Reproducibility gaps
Only summary statistics are pooled; individual-level data stay with each biobank (access terms in Supplementary Table 1). PRS weights are only promised for future PGS Catalog deposit; software versions and compute resources unreported.
Limitations (6)
  • Ancestry labels derive from global reference panels, but GBMI is not globally representative: most AFR and AMR participants come from US biobanks, so continental African and Central/South American ancestries are poorly captured. · stated by the authors
  • Effect sizes diverged for smaller, multi-ancestry or South Asian-including biobanks (BioMe, BioVU, UCLA, GNH) for gout, HF, VTE and POAG; POAG is also heterogeneous because self-report (Lifelines, TWB, BBJ) admits other glaucoma types. · stated by the authors
  • Gene prioritization methods agreed poorly, and TWAS was confined to European ancestry because GTEx has few non-European eQTL samples; the authors call for more robust in silico pipelines to nominate genes. · stated by the authors
  • Whether sex-differential signals (e.g., CTDP1/KCNG2 for asthma, ALDH2 and ALDH16A1 for gout) reflect pleiotropy, environmental exposures such as smoking or alcohol, or gene-environment interaction was not resolved. · stated by the authors
  • Liability-scale heritability assumes population prevalence equals pooled biobank prevalence, yet hospital-based biobanks had higher prevalence for six of 13 diseases, so these estimates may be biased. · noted in this summary
  • Improved polygenic risk prediction is asserted, but this text gives no R^2 values or confidence intervals, deferring to Supplementary Figure 12 and a companion study (Wang et al.). · noted in this summary

Read this if you design multi-biobank, multi-ancestry GWAS meta-analyses with EHR-harmonized phenotypes, or need GBMI summary statistics for asthma, gout, POAG, VTE, COPD or related endpoints.

Citations
437
FWCI
53.74
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISBhattachary '22Karczewski '24Sun '21Smith '22Li '23Troubat '24
74% semantic matchGenetic & Genomic Medicine
Arjun Bhattacharya, Jibril B Hirbo, Dan Zhou+7 more·Cell Genomics·Oct 1, 2022· 36 citations