A Bayesian nonparametric model to cluster SNPs data is proposed. A efficient Gibbs sampler is designed and a model based genetic score is computed. Association between groups of SNPs and a disease is assessed via the genetic score. The method is applied to a Crohn's disease study.