IMPUTE 5

IMPUTE 5 performs genotype imputation to predict unobserved genotypes using large reference panels of haplotypes for improved accuracy and computational efficiency.

Key Features:

• Positional Burrows Wheeler Transform (PBWT): Implements PBWT to rapidly identify locally best-matching haplotypes and long identical-by-state segments at genotyped markers.
• Per-individual haplotype selection: Selects custom subsets of reference haplotypes tailored for each individual's imputation and uses them as conditioning states within the IMPUTE model.
• Scalability: Demonstrates sub-linear scaling with reference panel size, with simulations showing less than a twofold increase in computation time when increasing from 10,000 to 1,000,000 haplotypes for a fixed set of imputed markers.
• Performance versus other methods: Empirically up to 30× faster than MINIMAC4 and 3× faster than BEAGLE5.1 while consuming less memory on the Haplotype Reference Consortium (HRC) reference panel (~65,000 haplotypes).
• Accuracy for low-frequency variants: Optimizes accuracy for imputing markers with low minor allele frequencies by conditioning on locally best-matching haplotypes.
• Reference panel subset utilization: Effectively uses a smaller subset of reference haplotypes as panel size grows, reducing computational demands for large-scale panels.
• Predecessor principles: Builds on principles from IMPUTE2 by refining haplotype conditioning strategies for improved accuracy and efficiency.

Scientific Applications:

• Genotype imputation: Predicts unobserved genotypes in study samples using large reference haplotype panels.
• Imputation of low minor allele frequency markers: Improves recovery of rare and low-frequency variant genotypes through tailored haplotype conditioning.
• Large-scale genomic studies: Enables imputation workflows for studies using large reference panels such as the Haplotype Reference Consortium (HRC) or simulated panels up to millions of haplotypes.
• Downstream population and association analyses: Provides imputed genotype data suitable for population genetics analyses and genome-wide association studies that require both accuracy and computational efficiency.

Methodology:

Implements the Positional Burrows Wheeler Transform (PBWT) to identify locally best-matching haplotypes and long identical-by-state segments at genotyped markers, selects optimal haplotypes per individual, and uses them as conditioning states within the IMPUTE statistical model.