Improving Imputation Quality in BEAGLE for Crop and Livestock Data

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Imputation is crucial for genetic study preprocessing.
• Existing imputation algorithms are often optimized for human genetics.
• Non-human genetic datasets present unique challenges due to varying diversity and structure.

Purpose of the Study:

• Evaluate BEAGLE imputation software versions on maize and chicken genetic data.
• Identify key parameters for optimizing imputation accuracy in non-human species.
• Compare the performance of different reference genomes and panel compositions.

Main Methods:

• Tested BEAGLE versions 5.0 and 5.1, alongside earlier versions.
• Tuned parameters including effective population size (ne) and haplotype cluster structure.
• Utilized European maize landraces, a commercial line, and a chicken diversity panel.
• Compared imputation accuracy using different reference genomes (flint vs. B73) and panel compositions.

Main Results:

• BEAGLE 5.0 showed superior phasing and imputation performance compared to 5.1 and earlier versions.
• Tuning the effective population size (ne) parameter reduced error rates by up to 98.5%.
• Using a flint reference genome halved high error rates in maize datasets.
• Excluding distant individuals from the chicken reference panel reduced average error rates by 8.5%.

Conclusions:

• BEAGLE parameter tuning is essential for accurate imputation in diverse genetic datasets.
• Effective population size and reference panel selection are critical factors.
• Optimizing imputation involves balancing genetic diversity representation with noise reduction.