Fig. 6

Summary of the five steps followed by the haplotype-based approach. The training data set was split and genotypes within each subset were independently phased using Shapeit (first step). From one half of the training data set, mAssocTest_2G was used to estimate the association between haplotypes (lengths considered were 2 to 4) and phenotype (second step). Different p-value thresholds were used to select the input variables (comprising 2 to 4 SNP-length haplotypes) (third step). Models of haplotype risk predictors were built using the second half of the training data set (fourth step). Individual risk was assessed by combining the two genome-wide haplotypes each individual has (fifth step)