Using Random Forests to scan for loci involved in oligogenetic disease

Peter Kraft
12 Dec 2002

Random Forests [L Breiman, Machine Learning 45:5-32] 
are a novel supervised learning method well-suited 
for data with many covariates but relatively few 
observations. The fitting procedure produces 
estimates of prediction error, variable importance, 
and observation proximity. Random forests can also be 
used in an unsupervised context to produce a matrix 
of intrinsic proximities among observations.

I review the basics of Random Forests and present an 
application in genetic epidemiology: detecting 
interactions among a few loci from a large set of 
(mostly inert) candidate loci in a simulated case-
control study.