The classical MHC locus encompasses approximately 3.6 megabase pairs on 6p21.3 and is divided into three subregions: the telomeric class I, class III, and the centromeric class II regions. It has been recently established by the evidence that both linkage disequilibrium and MHCrelated genes exist outside the classically defined locus. Genome-wide association study demonstrates that SNPs in the MHC region are strongly associated with psoriasis in different populations. Several SNPs are associated with psoriasis, but it is still unknown how many independent SNPs located AMN107 641571-10-0 within the MHC region contribute to the risk of psoriasis. The development of psoriasis is believed to involve a major locus PSOR1 in the MHC region and likely be in conjunction with multiple non-MHC loci with common alleles. Since MHC loci have been strongly associated with the development of psoriasis, identification of non-MHC loci associated with psoriasis may have been hindered by likely occurrence of genetic heterogeneity. In addition, a possible reason for the erratic replications of genetic association findings could be that the large effect of the PSORS1 locus may affect the effect of other loci involved in psoriasis. Therefore it is necessary to examine the genetic loci associated with psoriasis conditioning on the effect of the PSOR1 locus. Because of the extensive LD existing between the SNPs within MHC, identification of genetic variants to be associated with human disease is a challenging task. Routine haplotype analysis has a limited role in identifying independent SNPs in such a large linkage disequilibrium block within MHC. Conditional analysis approach adjusting for one top association signals from MHC have been used to search for other independent associations under an additive model. Since a number of association signals are often seen in the MHC region, selections of the top associated SNP for a conditional analysis can vary, consequently may lead to different results. In this study, we have employed a sophisticated approach to search for independent association signals within the MHC region. We first determined the variable importance of each SNP in the MHC region using both RandomForest algorithm and single SNP association, and then used each of the most important SNPs as a starting SNP to build a multiple regression model from more than 3,000 SNPs within the MHC. In the four regression models we built, two loci in HLA-C/HLA-B and HLA-DQA2 consistently appear in all models, and more importantly, rs9468925 in HLA-C.