henhaobio
Sign in

bioRxivGeneticsPreprint

Genome-wide genetic data on ~500,000 UK Biobank participants

Clare BycroftiD, Colin Freeman, Desislava Petkova, Gavin BandiD, Lloyd T. ElliottiD, Kevin SharpiD, Allan MotyeriD, Damjan VukceviciD, Olivier DelaneauiD, Jared O’ConnelliD, Adrián CortésiD, Samantha WelshiD, Gil McVeaniD, Stephen LeslieiD, Peter DonnellyiD, Jonathan MarchiniiD

PDFDOI

In-depth summary

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 1 (20 Jul 2017)

Quality-controlled genotypes for 488,377 UK Biobank participants at 805,426 markers, imputed to ~96 million variants, plus ancestry PCs, kinship (30.3% have close relatives enrolled) and HLA calls.

UK Biobank links rich health data on ~500,000 people to genome-wide genotypes, but its scale, batch design and ancestral diversity pose QC challenges, with population structure undermining standard checks such as Hardy-Weinberg tests. Compared with the ~150,000-person interim release, this full release uses an improved calling pipeline, the larger HRC imputation panel and a more compact BGEN format, and it ships ancestry-aware QC flags, relatedness and principal-component annotations and classical HLA calls.

Key results

  1. ShownThe quality-controlled genotype data are clean: few calls were blanked, duplicate samples agree almost perfectly, and allele frequencies track exome sequencing.

    488,377 samples × 805,426 markers released; 0.0097 of calls set to missing; 968 poor-quality samples (0.002); 588 duplicate pairs share 99.87% of calls on average (minimum 99.39%); allele frequencies vs ExAC at 91,298 markers give r2 = 0.94 in European-ancestry samples; 84% of rare markers pass QC in every batch. · 2.1.6 Summary of genotype data quality; 2.1.5 Sample-based QC

  2. ShownCryptic relatedness is common: close to a third of participants have a third-degree or closer relative in the cohort, with about twice the expected number of sibling pairs.

    KING kinship across all pairs: 147,731 participants (30.3%) in 107,162 related pairs, including 22,666 sibling pairs and 1,066 trios; PLINK re-inference on a subset confirmed 100% of twin, parent–offspring and sibling pairs and 99.9% of all pairs. The authors suggest volunteer and assessment-centre sampling bias as causes. · 2.2.3 Cryptic relatedness; 2.2.4 Trios and extended families

  3. ShownPhasing is accurate and imputation scaled to the whole cohort, raising the number of testable variants more than 100-fold.

    SHAPEIT3 switch error on chromosome 20 against family-derived haplotypes: median 0.229% in 696 British trios and 0.234% in 397 trios without third-degree relatives. IMPUTE4 with HRC plus UK10K/1000 Genomes panels took ~10 min per sample and yielded 92,693,895 autosomal variants in 487,442 people (~96 million variants in the abstract). · 2.3.1 Pre-Phasing; 2.3.2 Reference panel used for imputation; 2.3.3 Imputation; Abstract

  4. ShownClassical alleles at 11 HLA genes were imputed accurately enough to recover known disease associations.

    Five-fold cross-validation in the reference panel: four-digit accuracy above 93.9% at all 11 loci for European ancestry, above 96.1% when posterior > 0.70 (call rates above 95.1%). In 409,724 white British participants, top HLA alleles for 11 immune-mediated diseases matched previous reports, and IMSGC multiple sclerosis associations replicated. · 2.4 Imputation of classical HLA alleles

  5. SuggestedA height GWAS reproduced the known association landscape with more power than the largest meta-analysis that excludes UK Biobank.

    343,321 unrelated white British participants vs GIANT's 253,288 Europeans (2014): similar association patterns, many loci reaching p < 5x10-8 only in UK Biobank, and every GWAS-catalogue height region (as of Feb 2017) genome-wide significant or close; the lead chromosome 2 marker was imputed. The comparison is visual rather than quantified. · 2.5 GWAS for standing height

Methods, data and reproducibility

Samples were typed on two closely related Affymetrix Axiom arrays and called in 106 batches; marker QC (batch, plate, HWE, sex, array and replicate tests in a European-ancestry subset) blanked failing batches or dropped markers, while PCA-adjusted sample QC mostly flagged, not removed, samples. Genotypes were phased with SHAPEIT3, imputed with IMPUTE4 (HRC and UK10K/1000 Genomes panels) and HLA-typed with HLA*IMP:02.

  • UK Biobank participants genotyped on the UK BiLEVE Axiom (807,411 markers) and UK Biobank Axiom (825,927 markers) arrays · human · n = 488,377 released (49,950 UK BiLEVE; 438,427 UK Biobank Axiom) at 805,426 markers
  • European-ancestry subset for marker QC, found by projection onto PCs of 1000 Genomes CEU, YRI, CHB and JPT · human · n = 463,844
  • White British ancestry subset defined by self-report plus PCA · human · n = 409,728 (84%)
  • Imputation reference panels: Haplotype Reference Consortium (HRC, 39,235,157 SNPs) and merged UK10K + 1000 Genomes Phase 3 (87,696,888 markers) · human · n = 64,976 haplotypes (HRC); 12,570 haplotypes (UK10K + 1000 Genomes)
  • In-cohort trios used as a phasing truth set on chromosome 20 · human · n = 696 trios (397 without third-degree relatives)
  • ExAC exome allele frequencies used for comparison · human · n = 91,298 overlapping markers
  • Height GWAS sample and the GIANT 2014 height meta-analysis used for comparison · human · n = 343,321 (UK Biobank); 253,288 (GIANT)
  • HLA association analysis on self-reported disease codes in the white British subset · human · n = 409,724 individuals; 446 disease codes, 11 immune-mediated diseases
Methods & tools
Affymetrix custom genotype calling pipeline, fastPCA, KING, PLINK, SHAPEIT3, hapfuse, IMPUTE4, HLA*IMP:02, BGEN v1.2, BGENIX, BGENIE, QCTOOL v2
Reagents & environment
Named software: SHAPEIT3, IMPUTE4 (same HMM as IMPUTE2), HLA*IMP:02, KING, the BGEN library with BGENIX indexing and QCTOOL v2; BGENIE uses Eigen and OpenMP. No containers or further version details are given.
Experimental conditions
SHAPEIT3 ran in 15,000-marker chunks with 4 cores per job; IMPUTE4 used ~50,000-marker chunks and 5,000 samples per job, about 10 min of processing per sample genome-wide. The full imputed BGEN files take 2.1Tb (interim release: 1.3Tb).
Reproducibility gaps
Genotype, imputed and HLA data are available only through approved application to UK Biobank on a cost-recovery basis; a URL list for the software is given but the links are not in this text.
Limitations (6)
  • Very rare markers are hard to call on arrays: heterozygotes can blur into the major homozygote cluster, rare markers fail QC more often, and very rare ones tend to show lower MAF than in ExAC, so the authors advise visual checks of cluster plots. · stated by the authors
  • The white British subset may still carry subnational fine-scale structure, because methods able to detect it did not scale to this cohort at the time. · stated by the authors
  • Marker QC thresholds fail only strongly deviating markers, and poor-quality or putatively aneuploid samples are listed rather than removed (sex mismatches must be found by comparing reported and inferred sex), so further filtering is left to users. · stated by the authors
  • Array calling allows only haploid or diploid X/Y genotypes, so the 652 (0.134%) people with putative sex-chromosome aneuploidies may have compromised calls on those chromosomes. · stated by the authors
  • HLA imputation accuracy was estimated by cross-validation within the reference panel, not against directly typed UK Biobank samples. · noted in this summary
  • Several checks are qualitative or unshown: the height comparison with GIANT is visual, and IMPUTE4 matching IMPUTE2 and the posterior-filtered HLA analysis rest on data not shown. · noted in this summary

Read this if you analyse UK Biobank genotypes or imputed data and need to know which QC flags, ancestry subsets, kinship estimates and file formats come with the release, and how they were derived.

Citations
623

Abstract

The UK Biobank project is a large prospective cohort study of ~500,000 individuals from across the United Kingdom, aged between 40-69 at recruitment. A rich variety of phenotypic and health-related information is available on each participant, making the resource unprecedented in its size and scope. Here we describe the genome-wide genotype data (~805,000 markers) collected on all individuals in the cohort and its quality control procedures. Genotype data on this scale offers novel opportunities for assessing quality issues, although the wide range of ancestries of the individuals in the cohort also creates particular challenges. We also conducted a set of analyses that reveal properties of the genetic data - such as population structure and relatedness - that can be important for downstream analyses. In addition, we phased and imputed genotypes into the dataset, using computationally efficient methods combined with the Haplotype Reference Consortium (HRC) and UK10K haplotype resource. This increases the number of testable variants by over 100-fold to ~96 million variants. We also imputed classical allelic variation at 11 human leukocyte antigen (HLA) genes, and as a quality control check of this imputation, we replicate signals of known associations between HLA alleles and many common diseases. We describe tools that allow efficient genome-wide association studies (GWAS) of multiple traits and fast phenome-wide association studies (PheWAS), which work together with a new compressed file format that has been used to distribute the dataset. As a further check of the genotyped and imputed datasets, we performed a test-case genome-wide association scan on a well-studied human trait, standing height.

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
THISLi '23Halldórsson '22Canela‐Xand '18Shi '24Karczewski '24Lillo '20
72% semantic matchGenetic & Genomic Medicine
Shuwei Li, Keren Carss, Bjarni V. Halldórsson+1 more·medRxiv·Dec 6, 2023· 72 citations