Predictive modeling in case-control single-nucleotide polymorphism studies in the presence of population stratification: a case study using Genetic Analysis Workshop 16 Problem 1 dataset.

ABSTRACT: In this paper, we apply the gradient-boosting machine predictive model to the rheumatoid arthritis data for predicting the case-control status. QQ-plot suggests severe population stratification. In univariate genome-wide association studies, a correction factor for ethnicity confounding can be derived. Here we propose a novel strategy to deal with population stratification in the context of multivariate predictive modeling. We address the problem by clustering the subjects on the axes of genetic variations, and building a predictive model separately in each cluster. This allows us to control ethnicity without explicitly including it in the model, which could marginalize the genetic signal we are trying to discover. Clustering not only leads to more similar ethnicity groups but also, as our

SUBMITTER: Arshadi N 

PROVIDER: S-EPMC2795961 | biostudies-literature | 2009 Dec

REPOSITORIES: biostudies-literature