Preliminary analysis of a KIR haplotype estimation algorithm: a simulation study.

The analysis of Killer cell immunoglobulin-like receptors (KIRs) in terms of haplotypes have only been done through genotyping numerous and selected families. Consequently and schematically, KIR haplotypes have been roughly described by two groups (A and B) based on their gene contents. No further KIR adapted methods have been applied to the estimation of haplotype frequencies using unrelated data. We propose here a maximum likelihood (ML) estimation of KIR haplotype frequencies. ML estimation was developed as an extension of those successfully applied to human leukocyte antigen (HLA) data including the handling of missing values and HLA nomenclature. It has been implemented using an adapted Expectation Masimisation algorithm. KIR types on 11 loci in more than 40 Irish families were used to validate the method in a simulation study. Estimated haplotype frequencies are compared to the phase known. Various allele or gene frequency estimation methods were also compared. We demonstrated the interest and reliability of the haplotype method and underline the effect of the sample size on the quality of the estimation. The ML haplotype method also provides by collapsing more accurate estimation of allele or gene frequencies in population. Such an algorithm opens new perspectives in the analysis of KIR genotypes. Large sample size studies are required using phase-known data and/or simulations. It would allow a genotype-based approach to explore the KIR gene haplotype diversity. The haplotype frequencies may be used to compare populations.