Imputation-Based HLA Typing with GWAS SNPs.

Area of Science:

• Genetics
• Immunogenetics
• Bioinformatics

Background:

• Human leukocyte antigen (HLA) typing is crucial for immune system research and clinical applications.
• Major histocompatibility complex (MHC) region exhibits strong haplotype structure, enabling imputation methods.
• Genome-wide association studies (GWAS) often generate dense single nucleotide polymorphism (SNP) data suitable for HLA imputation.

Purpose of the Study:

• To describe the HIBAG (imputation method with attribute bagging) workflow for inferring HLA classical alleles from SNP data.
• To demonstrate the utility of HIBAG for genotype imputation and model training using public 1000 Genomes project data.

Main Methods:

• Utilized SNP genotypes from the 1000 Genomes project for HLA imputation.
• Employed HIBAG, an imputation method incorporating attribute bagging.
• Developed a GPU-accelerated implementation for significantly faster model building.

Main Results:

• Successfully inferred HLA classical alleles from SNP data using the HIBAG workflow.
• Demonstrated two use cases: genotype imputation within a GWAS context and de novo model training.
• The GPU implementation achieved hundreds of times speedup compared to single-threaded computation.

Conclusions:

• HIBAG provides an efficient and cost-effective approach for HLA imputation from SNP data.
• The method facilitates HLA analysis on existing GWAS datasets without additional genotyping costs.
• Accelerated model building with GPU support enhances the practicality of HIBAG for large-scale studies.