Mixture models detect large effect QTL better than GBLUP and result in more accurate and persistent predictions

Table 2 Accuracy of prediction of phenotypes in generation 8 based on training in all seven previous generations (G1-G7) or any one of the previous 7 generations for pedigree BLUP (PBLUP) and four genomic methods (BayesB, BayesC, BayesC0 and GBLUP)

	N	Method of training with validation in generation 8
Training		PBLUP^a	BayesB^b	BayesC^b	BayesC0	GBLUP^c
G1-G7	1,814	0.27	0.60	0.57	0.50	0.50
G1	295	0.10	0.35	0.33	0.19	0.20
G2	323	0.00	0.36	0.36	0.24	0.24
G3	294	0.02	0.49	0.45	0.28	0.28
G4	360	−0.07	0.40	0.30	0.12	0.12
G5	290	−0.01	0.47	0.38	0.32	0.32
G6	252	0.19	0.49	0.44	0.33	0.34
G7	295	0.22	0.48	0.45	0.37	0.37
Average^d		0.06	0.43	0.39	0.26	0.27

^aPedigree-based BLUP
^bMixture models assumed the fraction of SNPs with 0 effect (π) of 0.99
^cGBLUP was fitted as BayesC0 with genetic and residual scale factors having 100° of freedom
^dAverage of the 7 individual generation results