. 2017 Feb 1;49:17. doi: 10.1186/s12711-017-0293-6

Table 3.

Accuracy of genomic prediction for BCWD resistance with BayesB using progeny testing families in five GS schemes

GS scheme	Family		Training size	Training–testing relationship^b	$π$ ^c	SNPs^d	DAYS^e		STATUS^e
GS scheme	Number	Size	Training size	Training–testing relationship^b	$π$ ^c	SNPs^d	$P A_{G E B V}$ ^f	$B i a s_{G E B V}$ ^g	$P A_{G E B V}$ ^f	$B i a s_{G E B V}$ ^g
1	50	20-40^a	1473	0.66	0.97	1069	0.71	1.16	0.71	1.01
2	50	20	991	0.50	0.98	713	0.67	1.55	0.67	1.51
3	25	40	979	1.00	0.98	713	0.69	1.26	0.72	1.23
4	25	20	497	1.00	0.987	463	0.53	1.37	0.61	1.66
5	25	20	494	0.00	0.987	463	0.25	3.33	0.22	5.08

A sample of 193 testing fish (from total n = 930 testing fish) were inter-mated to develop 138 progeny testing families (PTF). After disease evaluation of progeny from the 138 PTF (n = 9968), we estimated the mean progeny phenotype (MPP) for each PTF

^aIn scheme1, there were two groups of training families: (1) A set of 25 families with 40 offspring each that contributed fish to the testing sample; and (2) A set of 25 families with 20 offspring each that did not contribute fish to the testing sample

^bProportion of training fish that were full-sibs (FS) of testing fish: scheme 1 = 0.66 indicates that 66% of training fish were FS of testing fish; scheme 2 = 0.50 indicates that 50% of training fish were FS of testing fish; schemes 3 and 4 = 1.0 indicates that ALL training fish were FS of testing fish; and scheme 5 = 0.0 indicates that NONE of training fish were FS of testing fish (i.e., training and testing fish were sampled from different families)

^cBayesB method uses a mixture parameter $π$ that specifies the proportion of loci with zero effect, and the analyses included 35,636 effective SNPs

^dNumber of SNPs that are sampled as having non-zero effect $(1 - π)$ and fitted simultaneously in the multiple regression model

^eBacterial cold water disease (BCWD) resistance phenotypes: BCWD survival days (DAYS) and survival status (STATUS)

^fPredictive ability of GEBV $(P A_{G E B V})$ was defined as the correlation of MPP with mid-parent GEBV from each PTF: $P A_{G E B V} = C O R R (M P P, M i d p a r e n t G E B V)$

^gBias of GEBV $(B i a s_{G E B V})$ was defined as the regression coefficient of performance MPP on predicted mid-parent GEBV: $B i a s_{G E B V} = R E G R E S (M P P, M i d p a r e n t G E B V)$ . The predicted GEBV for STATUS estimated on the underlying scale of liability were transformed to the observed scale (probability of survival)