PONG 2.0: allele imputation for the killer cell immunoglobulin-like receptors

Killer cell immunoglobulin-like receptors (KIRs) are polymorphic immune regulators that modulate natural killer and T cell responses via interactions with human leukocyte antigen (HLA) class I ligands. High combinatorial diversity of KIR and HLA influences infection, autoimmunity, cancer, transplantation, and reproductive success. Although comprehensive KIR genotyping is achievable through targeted sequencing, complex genomic architecture hampers large-scale disease studies using genome wide data. Here, we introduce PONG2.0, a computational framework that accurately imputes high-resolution genotypes for all KIR that interact with HLA, directly from SNP-array data. We trained multi-ancestry models using matched SNP and KIR alleles from a subset of the 1000 Genomes Project (EUR = 187, AMR = 93, SAS = 102, AFR = 102, EAS = 102), achieving 92-99% overall accuracy. Validation against targeted sequencing of 267 independent samples confirmed robust per-locus concordance (92.1-97.7%). Population-level benchmarking of > 8000 individuals showed strong agreement with targeted sequencing (R2 up to 0.999; median deviation 0.6-6.8%), demonstrating reliable genotyping of samples that were independent from the model-building data. PONG2.0 is implemented as an open-source R package with pre-trained models, providing an efficient and scalable solution for KIR immunogenetics in large-scale biobanks and cohort studies. https://github.com/NormanLabUCD/PONG2.