IMPUTE2

IMPUTE2 performs genotype imputation and haplotype phasing by predicting unobserved genotypes from haplotype patterns in reference panels (e.g., 1000 Genomes Project) to support genetic association studies.

Key Features:

• Custom Reference Panel Selection: Uses local sequence similarity to select a custom reference panel for each study haplotype across genomic regions, enabling use of all available reference haplotypes.
• Flexible Modelling Framework: Integrates information across multiple reference panels within a flexible statistical modelling framework to improve imputation accuracy while remaining computationally feasible.
• Integration of Heterogeneous Reference Panels: Combines information from reference panels genotyped on different sets of SNPs to support imputation when reference panels differ in SNP content.
• Enhanced Accuracy Across Populations: Demonstrates accurate imputation across diverse human populations, including results reported using HapMap 3 and the Malaria Genetic Epidemiology Network (MalariaGEN).
• Performance with Large Reference Panels: Scales to reference panels containing thousands of chromosomes and reports overall error rates 15%–20% lower than the closest alternative methods at rare and common SNPs.

Scientific Applications:

• Genome-wide association studies (GWAS): Improves genotype density and accuracy in GWAS by imputing untyped variants from large reference datasets.
• Imputation-based studies in diverse populations: Enables imputation in samples from diverse genetic backgrounds, including African populations, by leveraging population-specific and multi-population reference panels.
• Next-generation association studies and large-scale sequencing integration: Supports use of very large reference panels and heterogeneous SNP sets typical of next-generation sequencing–based association analyses.

Methodology:

Predicts unobserved genotypes using statistical imputation based on haplotype patterns from reference panels, selects custom reference haplotypes per study haplotype via local sequence similarity, and integrates information across multiple reference panels and heterogeneous SNP sets within a flexible modelling framework.