A comparison of principal component regression and genomic REML for genomic prediction across populations

Table 6 Highest accuracies ¹ obtained for PCR models ² versus those obtained with the GREML model ³

Test	Trait	GREML	PCR_eigen	PCR_ss	SSPCR_eigen	SSPCR_ss
UK_1	Milk	0.26	0.44 (14)	0.36 (2)	0.45 (6)	0.23 (458)
	Fat	0.07	0.16 (776)	0.13 (3)	0.18 (1)	0.21 (219)
	Protein	−0.01	0.25 (14)	0.16 (1)	0.28 (7)	0.09 (1)
UK_2	Milk	0.10	0.16 (3)	0.15 (11)	0.19 (144)	0.15 (46)
	Fat	−0.02	0.08 (1061)	0.07 (751)	0.12 (593)	0.09 (1370)
	Protein	0.01	0.07 (1)	0.08 (2)	0.10 (151)	0.08 (233)
SWE	Milk	0.16	0.18 (1112)	0.16 (1060)	0.21 (365)	0.17 (344)
	Fat	0.09	0.22 (46)	0.10 (991)	0.16 (1425)	0.18 (871)
	Protein	0.06	0.11 (265)	0.08 (790)	0.25 (1424)	0.20 (1419)
IRL	Milk	0.06	0.15 (967)	0.12 (758)	0.14 (92)	0.19 (288)
	Fat	0.08	0.12 (954)	0.14 (572)	0.16 (790)	0.12 (965)
	Protein	0.04	0.12 (749)	0.09 (245)	0.16 (94)	0.13 (273)
NLD	Milk	0.16	0.21 (20)	0.17 (4)	0.19 (24)	0.17 (16)
	Fat	0.15	0.17 (794)	0.19 (400)	0.17 (585)	0.18 (78)
	Protein	0.13	0.17 (7)	0.16 (8)	0.17 (5)	0.17 (4)
Average	Milk	0.15	0.23 (423)	0.19 (367)	0.24 (126)	0.18 (230)
	Fat	0.07	0.15 (726)	0.13 (543)	0.16 (679)	0.16 (701)
	Protein	0.05	0.14 (207)	0.11 (209)	0.19 (336)	0.13 (386)
SD	Milk	0.08	0.12 (565)	0.10 (506)	0.12 (144)	0.03 (192)
	Fat	0.06	0.05 (398)	0.05 (373)	0.02 (511)	0.05 (540)
	Protein	0.05	0.07 (323)	0.04 (341)	0.07 (611)	0.05 (591)

Analyses were performed for three traits and five test populations.
¹Accuracies were calculated as Pearson correlations between predicted genomic breeding values and adjusted phenotypes; ²selection of PC was based either on the eigenvalues (eigen) or the regression sum of squares (ss); two different methods of applying principal component analysis, either separately for reference and test parts (PCR) or on the whole dataset (SSPCR), were compared; ³a REML based model with a genomic relationship matrix.

ISSN: 1297-9686