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

Table 4 Accuracy of predicting phenotype in each successive generation when using only the marker effect estimates from the top genomic 1 Mb window based on training in generation 1 for four genomic methods (BayesB, BayesC, BayesC0 and GBLUP)

Validation generation	Method of training in generation 1
Validation generation	BayesB^a	BayesC^a	BayesC0	GBLUP^b
G2	0.25	0.23	−0.06	−0.03
G3	0.23	0.19	0.08	0.08
G4	0.27	0.26	−0.03	−0.10
G5	0.29	0.28	−0.13	−0.15
G6	0.31	0.30	0.11	0.13
G7	0.30	0.29	−0.03	−0.02
G8	0.34	0.34	0.00	0.00
Average	0.28	0.27	−0.01	−0.01

^a Mixture models assumed the fraction of SNP with 0 effect (π) of 0.99
^b GBLUP was fitted as BayesC0 with genetic and residual scale factors having 100 df

ISSN: 2049-1891