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

Table 1 Comparison of four genomic methods (BayesB, BayesC, BayesC0 and GBLUP) for QTL detection using different generations (G1 to G7) of training data

	BayesB			BayesC			BayesC0		GBLUP
Training Data	chr_Mb^a	%Var^b	p >0^c	chr_Mb	%Var	p >0	chr_Mb	%Var	chr_Mb	%Var
G1-G7	4_78	23.7	1.00	4_78	14.4	1.00	4_77	0.62	4_77	0.62
G1	4_78	9.4	0.82	4_78	2.5	0.68	3_110	0.36	3_110	0.32
G2	4_78	6.1	0.73	4_78	6.1	0.73	Z_39	0.31	Z_39	0.32
G3	4_78	13.8	0.64	4_78	3.1	0.77	Z_9	0.32	Z_27	0.35
G4	4_78	22.7	0.98	4_78	4.1	0.90	Z_27	0.54	Z_27	0.58
G5	4_78	10.4	0.77	4_78	1.1	0.50	Z_27	0.35	Z_27	0.36
G6	4_78	7.1	0.77	3_42	3.2	0.93	3_110	0.38	3_110	0.31
G7	Z_23	4.7	0.77	Z_23	3.6	0.77	Z_23	0.32	Z_23	0.34

^aLocalization of the 1 Mb window that explained the largest amount of variance (chromosome_1 Mb window on that chromosome)
^bpercentage of variance explained by that window
^cproportion of models where this window accounted for more than 0 % of genetic variance

ISSN: 2049-1891