Multifactor Dimensionality Reduction for Detecting Haplotype-Haplotype Interaction

The multifactor dimensionality reduction (MDR) is a model-free approach that can identify SNP × SNP effects in a case-control study. In this study, we extended MDR to identify interactions among haplotypes (Hap-MDR). With Hap-MDR, multilocus haplotype genotypes were pooled into high-risk and low-risk groups, effectively reducing the haplotype genotypes predictors from N dimensions to one dimension. Here we also provided four model selection methods: 1) combining cross-validation and accuracy (ACC) to select model; 2) combing cross-validation and Mathew´s correlation coefficient (MCC) to select model; 3) combining cross-validation and Pearson chi-square to select model; 4) using Pearson chi-square to select model only. The effectiveness of the approach was demonstrated by extensive experimental studies using simulated data set which include 10 independent replicates. The results showed that our hap-MDR method was powerful. The second model selection method has the highest average cross-validation consistency C_average = 5.38 and the lowest permutation evaluating index N_average = 0. Compared with MDR, Hap-MDR can overcome influence of linkage disequilibrium (LD), and can improve the analysis efficiency.