Genomic prediction with epistasis models: on the marker-coding-dependent performance of the extended GBLUP and properties of the categorical epistasis model (CE)

Table 5 Predictive abilities on the wheat data when prior information is incorporated in the marker coding of EGBLUP. Predictive abilities when the coding for each interaction is determined based on records under different environmental conditions

	G-Env 1	G-Env 2	G-Env 3	G-Env 4
Environment 1	——	0.555 ± 0.007	0.559 ± 0.007	0.552 ± 0.007
Environment 2	0.503 ± 0.007	——	0.544 ± 0.006	0.514 ± 0.007
Environment 3	0.394 ± 0.008	0.430 ± 0.008	——	0.402 ± 0.008
Environment 4	0.500 ± 0.007	0.511 ± 0.006	0.513 ± 0.006	——

G-Env 1 means that the relationship matrix was constructed under the use of the data of Environment 1 (analogously for other environments; for a description of the construction of the matrices see section “Methods”). Bold numbers indicate predictive abilities higher than that of all previously used methods for this trait

ISSN: 1471-2105