P057 Measuring linkage disequilibrium with next-generation sequencing HLA typing data

Abstract

Aim To show how HLA typing data from next-generation sequencing (NGS) can dramatically improve our estimates of linkage disequilibrium associations between different loci. Method Given a set of haplotypes in which loci have been placed in phase, for each pair of alleles (Xi, Yi) in two loci X and Y, a 2 × 2 contingency table (a, b, c, d) is built with the following counts: a = CARD(Xi and Yi); b = CARD(X that is not Xi and Yi); c = CARD(Xi and Y that is not Yi), and d = CARD(Xi that is not Xi and Y that is not Yi), where CARD() is a function that returns the cardinality of a set. From these 2 × 2 contingency tables several measures can be calculated, like the classical measure of linkage disequilibrium delta prime, or ‘mutual information’ from information theory, or the ‘ Ochiai semidistance’ from geometry, or chi square from statistical theory, or the likelihood ratio, as well as various useful conditional probabilities. Several testing sets with loci in phase were used. Linkage disequilibrium calculations were made at the classical high-resolution level (4-digit resolution) and at the full resolution level (4-field resolution). NGS typing was done with the Mia Fora technology from Immucor. Results A set of 755 DRB1-DQB1 blocks was used to calculate linkage disequilibrium between Xi = DQA1 ∗ 01:02 and Yi = DRB1 ∗ 13:02 (4-digit res.) and also between Xi = DQA1 ∗ 01:02:01:04 and Yi = DRB1 ∗ 13:02:01 (4-field res.). The results for the first comparison (4-digit res.) were (a, b, c, d) = (70, 0, 196, 489); and for the second comparison (4-field res.) were (a, b, c, d) = (70, 0, 0, 685). While the conditional probability of having DRB1 ∗ 13:02 in the presence of DQA1 ∗ 01:02 (4-digit res.) is 26%, the conditional probability of having DRB1 ∗ 13:02:01 in the presence of DQA1 ∗ 01:02:01:04 (4-field res.) is 100%. The chi square went from 141 for 4-digit res. to 755 for 4-field res. This is just an example of many such findings which will be presented at ASHI 2016. Conclusions Using 4-field resolution data from NGS often brings linkage disequilibrium to a much stronger level, sometime to a total certainty. The most efficient way to present linkage disequilibrium data is, as detailed here, in the form of the cardinality of the logical possibilities as a 2 × 2 contingency table.