Evaluation of two methods for computational HLA haplotypes inference using a real dataset.

Area of Science:

• Population genetics
• Immunogenetics
• Computational biology

Background:

• High polymorphism of HLA haplotypes is valuable for population genetics and disease association studies.
• Existing computational methods for haplotype inference lack clear accuracy assessments.
• Direct comparison of leading methods is needed to determine superiority.

Purpose of the Study:

• To compare the accuracy of two leading computational haplotype inference methods: Expectation-Maximization (Arlequin V3.0) and Bayesian (PHASE V2.1.1).
• To evaluate performance using real HLA haplotypic data (A-B-Cw-DQB1-DRB1).

Main Methods:

• Direct counting of 122 HLA haplotypes.
• Accuracy assessment using Mean Squared Error (MSE), Similarity Index (IF), and Haplotype Identification Index (IH).
• Comparison of Arlequin V3.0 (EM-based) and PHASE V2.1.1 (Bayesian-based) algorithms.

Main Results:

• Neither method perfectly inferred all known haplotypes.
• Some differences in haplotype determination and frequency estimation accuracy were observed between the two methods.
• Increased confidence in haplotype assignment and frequency estimation when using low polymorphic sites present in multiple individuals.

Conclusions:

• The Bayesian algorithm (PHASE V2.1.1) demonstrated superior overall performance in haplotype construction and frequency calculation.
• Differences between the two methods were generally insubstantial.
• This study represents the first evaluation of statistical haplotype inference methods using real HLA region data.