Bayesian construction of perceptrons to predict phenotypes from 584K SNP data.
Luc Janss and Bert Kappen
In: PASCAL Computational Statistics Workshop, June 2009, London, UK.

Abstract

Introduction ● Genetic prediction still difficult ● But also done very simple, testing single predictors.... ● Here: Bayesian multivariate model building ● Based on “Variable Selection” George & McCullogh 1993 ● Constructs complete “perceptrons”: the simultaneous bitpattern for selected predictors ● Example: human data 632 phenotypes + orginally 710K SNPs (p/n >1000) ● > Using 3 chromosomes, 104K SNPs ● > Cross validation using 80:20 split

[Edit]