A Benchmark of Modern Statistical Phasing Methods

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Accurate haplotype phasing is essential for genomic analyses.
• Evaluating the performance of statistical phasing methods is challenging.
• Existing benchmarks may not fully capture real-world phasing accuracy.

Purpose of the Study:

• To benchmark and compare the accuracy of three leading statistical phasing methods: Beagle 5.4, SHAPEIT 4, and Eagle v2.4.1.
• To identify method-specific error patterns and their potential causes.
• To assess the utility of synthetic diploid data for phasing accuracy evaluation.

Main Methods:

• Generation of synthetic diploid genomes from male X chromosome sequences.
• Application of Beagle 5.4, SHAPEIT 4, and Eagle v2.4.1 to phased genomic data.
• Comparative analysis of phasing accuracy using switch and flip error rates.
• Validation using autosomal trio data.

Main Results:

• All tested methods exhibited correlated error rates, but with distinct error type frequencies.
• Eagle v2.4.1 showed a higher rate of switch errors; SHAPEIT 4 showed a higher rate of flip errors.
• Errors were enriched at CpG sites and rare variant sites, particularly flip errors, across diverse populations.

Conclusions:

• Synthetic diploid data provides a robust framework for evaluating phasing accuracy.
• Method-specific error profiles exist and are influenced by genomic context (CpG, rare variants).
• Understanding these error patterns is critical for accurate downstream genomic analyses and highlights potential biases in validation datasets.