MaCH-admix: genotype imputation for admixed populations

Area of Science:

• Genetics
• Bioinformatics
• Population Genetics

Background:

• Genotype imputation in admixed populations is complex due to intricate linkage disequilibrium (LD) patterns.
• Large reference panels, like the 1,000 Genomes Project, offer potential for improved imputation accuracy, particularly for admixed groups and rare variants.
• Effective reference panel selection is critical for maximizing the benefits of large reference datasets within genotype imputation frameworks.

Purpose of the Study:

• To develop and evaluate novel methods for constructing effective reference panels tailored to individual target samples for genotype imputation in admixed populations.
• To compare the performance of different reference panel construction strategies, including identity-by-state (IBS) based and ancestry-weighted approaches, within a hidden Markov model framework.
• To assess imputation quality for uncommon variants in diverse admixed populations using large-scale datasets.

Main Methods:

• Implemented and evaluated several reference panel construction methods, categorizing them into identity-by-state (IBS) and ancestry-weighted approaches.
• Utilized a hidden Markov model framework for genotype imputation, with methods specifically designed for individual target samples.
• Conducted performance evaluations on large cohorts of African Americans and Hispanic Americans from the Women's Health Initiative, comparing against established software like BEAGLE and IMPUTE2.

Main Results:

• The novel piecewise IBS method demonstrated consistently superior imputation quality compared to other evaluated methods and software when using large reference panels.
• A significant information gain of up to 5.1% was observed for uncommon variants using the piecewise IBS method, with highly statistically significant differences (P-value < 0.0001).
• Performance was assessed across various experimental conditions, including different reference panel sizes, target sample sizes, and varying levels of genomic LD.

Conclusions:

• The developed piecewise IBS method represents a significant advancement in genotype imputation for admixed populations, offering enhanced accuracy.
• This study provides the first comprehensive comparison of various sensible approaches for imputation in admixed populations, highlighting the effectiveness of tailored reference panel construction.
• The findings underscore the importance of optimized reference panel selection for improving imputation of rare variants in genetically diverse groups.