Fig. 5
Superiority of BLUP methods over Bayesian LASSO on real traits. Three BLUP methods (gBLUP, cBLUP, and sBLUP) were compared with Bayesian LASSO. Superiority was defined and calculated as the difference in prediction accuracy between each BLUP method and Bayesian LASSO. Each trait is presented as a dot and positioned according to its heritability and superiority value for the three BLUP methods. A dot is filled solid if a BLUP method is superior to Bayesian LASSO; otherwise, a dot is outlined. We evaluated 81, 41, and 35 real traits from Arabidopsis (a–c), mice (d–f) and maize (g–i), respectively. The Arabidopsis data contained 21 traits on flowering time that were classified as complex traits in a previous study (cite). The complex traits from Arabidopsis are colored red; the other traits (simple) are colored blue (a–c). We did not differentiate complexity for the traits in maize or mice. All traits are colored black in these two species (d–i). For most of the simple traits from Arabidopsis (blue dots), sBLUP is superior to Bayesian LASSO (c). For the mice traits with low heritability (<40%), cBLUP is superior to Bayesian LASSO (e). In maize, Bayesian LASSO is superior to gBLUP for most of the traits, but not compared with cBLUP or sBLUP(g)
