Imputing HLA-G high-resolution alleles and regulatory haplotypes from exomes and SNP array data

HLA-G encodes an immune checkpoint molecule with restricted expression in immune-privileged tissues and pathological conditions. It exhibits limited coding diversity but substantial regulatory-region variation influencing expression levels. Strong linkage disequilibrium across HLA-G creates a structured genetic architecture in which regulatory and coding variants co-segregate into well-defined haplotypes, enabling the imputation of complete HLA-G haplotypes from partial genomic data. We developed imputation models to predict HLA-G 4-field alleles, promoter, and 3'UTR haplotypes from whole-exome sequencing and SNP array data using HIBAG. Multi-ethnic reference panels were constructed from 5,347 individuals from three diverse cohorts (1000 Genomes, Human Genome Diversity Project, and Brazilian SABE cohort). Models were validated through cross-validation and independent datasets. Exome-based imputation achieved high accuracy (>99%) for common alleles (frequency > 1%), with mean posterior probabilities exceeding 0.95. SNP array-based models showed slightly lower but still robust performance (>95% accuracy). Our approach enables simultaneous prediction of coding and regulatory sequences, providing comprehensive functional information from datasets that do not capture the complete HLA-G diversity. These models facilitate HLA-G analysis in widely available genomic datasets lacking introns and regulatory regions (tumor exomes, SNP arrays), enabling investigation of HLA-G's role in immune regulation, transplantation, cancer, and pregnancy complications without full-gene sequencing.