Comparison of different imputation methods from low- to high-density panels using Chinese Holstein cattle

Area of Science:

• Animal Genetics
• Bioinformatics
• Genomic Selection

Background:

• Imputing high-density genotypes from lower-density platforms offers a cost-effective strategy for enhancing whole-genome selection programs in livestock.
• Accurate genotype imputation is crucial for maximizing the genetic gain in breeding programs.

Purpose of the Study:

• To evaluate and compare the imputation accuracy and efficiency of three widely used algorithms: fastPHASE, BEAGLE, and findhap.
• To assess the impact of reference population size and SNP density on imputation performance in Chinese Holstein cattle.

Main Methods:

• Utilized Illumina BovineSNP50 genotypes from 2108 Chinese Holstein cattle, divided into reference and test populations.
• Simulated varying SNP densities by masking 20% to 95% of single-nucleotide polymorphisms (SNPs) on three bovine chromosomes (BTA1, 16, 28).
• Compared imputation accuracies and computational speeds of fastPHASE, BEAGLE, and findhap under different scenarios.

Main Results:

• BEAGLE exhibited the highest imputation accuracy (>90%) and robustness across various SNP densities and reference population sizes.
• fastPHASE performance was significantly affected by the proportion of masked SNPs, particularly at high masking rates.
• findhap offered the fastest computation but achieved lower accuracy than BEAGLE, though superior to fastPHASE.

Conclusions:

• BEAGLE is the most reliable algorithm for imputing genotypes from low- to high-density platforms, balancing accuracy and computational demands.
• The choice of imputation algorithm is critical, with BEAGLE recommended for robust genomic selection in cattle.
• Reference population size positively influences imputation accuracy for BEAGLE and findhap, but not fastPHASE.