HIBAG

HIBAG imputes human leukocyte antigen (HLA) alleles from dense single-nucleotide polymorphism (SNP) genotype data to enable high-resolution HLA typing for genetic and immunogenetic studies by leveraging extended haplotype structure within the major histocompatibility complex (MHC).

Key Features:

• Attribute bagging approach: Employs attribute bagging with bootstrap aggregating and random variable selection to construct an ensemble of classifiers and average HLA-type posterior probabilities for prediction.
• Use of MHC haplotype structure: Leverages the extended haplotype structure within the MHC to predict HLA alleles from dense SNP genotypes.
• Training parameter options: Can utilize published parameter estimates in place of individual-level training datasets.
• Input data formats: Operates on dense SNP genotypes from genome-wide SNP panels, using SNP markers common across several Illumina platforms.
• Performance assessment: Demonstrated prediction accuracies of 92.2%–98.1% for HLA-A, B, C, DRB1, and DQB1 using a 2668-subject European training set and independent validation in the British 1958 birth cohort (~1000 subjects).
• Comparative evaluation: Performed competitively in comparative analyses against methods such as HLA*IMP and BEAGLE.
• Implementation: Implemented as an R package and includes pre-fit classifiers for European, Asian, Hispanic, and African ancestries.

Scientific Applications:

• Disease association studies: Imputes HLA alleles for association analyses of MHC-linked diseases and immune-related traits.
• Adverse drug reaction research: Facilitates investigation of HLA-associated drug hypersensitivity and adverse drug reactions.
• Population genetics: Enables large-scale population studies of HLA allele distributions across ancestries using SNP genotype datasets.
• Genetic predisposition analyses: Supports studies of genetic predispositions and immunogenetic risk factors where direct HLA typing is impractical.

Methodology:

Constructs ensembles of classifiers via attribute bagging (bootstrap samples with random variable selection) and averages HLA-type posterior probabilities; can use published parameter estimates; validated using a 2668-subject European training set and independent validation in the British 1958 birth cohort (~1000 subjects) with SNP markers common across several Illumina platforms; implemented in an R package with pre-fit ancestry-specific classifiers.