Fig. 2
From: Genome-wide prediction using Bayesian additive regression trees
Manhattan plots of the penalized regression coefficients from the LASSO, BLASSO, GBLUP and RKHS methods, VIMP (percent decrease in MSE) for RF, VIP (average number of node splits per iteration) for BART from the analyses of the QTLMAS2010 dataset with non-additive genetic effects added on chromosome 5. Dotted lines delimit chromosomes. The dominant locus is indicated by a red square, the over-dominant locus by a green upper triangle, the under-dominant locus by a cyan lower triangle, and the epistatic loci by blue diamonds