UK Biobank imputation pipelines

Genotype imputation pipelines for the UK Biobank Research Analysis Platform

Introduction

Genotype imputation is a computational technique for estimating missing genotypes in SNP array data, using a reference panel of haplotypes. This approach extends to low-coverage whole genome sequencing data, aiding in filling missing genotypes or enhancing uncertain genotype calls from sequencing reads.

For both SNP array and low-coverage whole genome sequencing data, we've created two distinct pipelines using the UK Biobank reference panel (>200,000 samples; 700M variants) for genotype imputation. To ensure cost-effective implementation, we leverage efficient state-of-the-art tools, including IMPUTE5 (Rubinacci et al., 2020) for SNP array imputation and GLIMPSE2 (Rubinacci et al., 2023) for low-coverage WGS imputation.

Our pipelines can take input from a multi-sample VCF/BCF file with SNP array genotypes or a set of low-coverage BAM/CRAM files. Using the UK Biobank reference panel, the pipeline executes imputation through applets and dx command jobs, tailor-made for the UKB RAP. At the end of each imputation pipeline, a single multi-sample BCF file is generated per chromosome, encompassing genotype posteriors, dosages, and phased best-guess genotypes. Further outputs like haploid dosages can be obtained by specifying appropriate options in the imputation software.