Comprehensive Assessment of Genotype Imputation Performance.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Genotype imputation is crucial for enhancing genome-wide association studies (GWAS), meta-analyses, and fine-mapping by estimating missing genotypes.
• Factors influencing imputation performance include software choice, reference panel selection, sample size, and SNP density or sequencing coverage.
• Systematic evaluation of popular imputation software is needed to guide future genetic studies.

Approach:

• Evaluated four popular genotype imputation software: Beagle4.1, IMPUTE2, MACH+Minimac3, and SHAPEIT2+IMPUTE2.
• Utilized East Asian ancestry test samples and the 1000 Genomes Project reference panel for performance assessment.
• Assessed imputation performance across varying SNP densities, sample sizes, sequencing depths, and for low minor allele frequency SNPs.

Key Points:

• IMPUTE2 achieved the highest accuracy (99.18%), closely followed by SHAPEIT2+IMPUTE2 (99.08%), Beagle4.1 (98.94%), and MACH+Minimac3 (98.51%).
• A minimum SNP density of > 200/Mb is required for good and stable imputation quality.
• Imputation accuracy from sequencing with 15× depth can be largely achieved by imputing 4× depth sequencing data using the 1000 Genomes Project reference panel.
• All software showed weaker performance in regions with low minor allele frequency SNPs due to reference or software bias.

Conclusions:

• IMPUTE2 and Beagle4.1 showed minor influence from study sample size on imputation accuracy.
• Reference panel contribution to imputation performance is software-dependent.
• Future advancements in reference panels and algorithms are necessary to address challenges in low minor allele frequency SNP imputation.