A screening methodology based on random forests to improve the detection of gene-gene interactions
Résumé
The search for susceptibility loci in gene-gene interactions imposes a methodological and computational challenge for statisticians due to the large dimensionality inherent to the modelling of gene-gene interactions or epistasis. In an era where genome-wide scans have become relatively common, new powerful methods are required to handle the huge amount of feasible gene-gene interactions and to weed out the false positives and negatives from these results. One solution to the dimensionality problem is to reduce the data by preliminary screening of markers to select the best candidates for further analysis. Ideally, this screening step is statistically independent of the testing phase. Initially developed for small numbers of markers, the Multifactor Dimensionality Reduction method is a nonparametric, model-free data reduction technique to associate sets of markers with optimal predictive properties to disease. In this study, we examine the power of Multifactor Dimensionality Reduction in larger datasets and compare it to other approaches that are able to identify gene-gene interactions. Under a variety of interaction models (purely and not purely epistatic), we use a Random Forests -based pre-screening method, before executing the Multifactor Dimensionality Reduction, to improve its performance. We find that the power of Multifactor Dimensionality Reduction increases when noisy SNPs are first removed by creating a collection of candidate markers with Random Forests. We validate our technique by extensive simulation studies and by application to asthma data from the ECRHS II study.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...