Table 4.

Performance comparison of deepGBLUP with the other genomic prediction methods on the Korean native cattle dataset across different traits and training sizes

Train size	Method	CWT	BF	EMA	MS
9000	GBLUP	0.729 ± 0.015	0.647 ± 0.009	0.726 ± 0.017	0.670 ± 0.014
	DGBLUP	0.731 ± 0.016	0.639 ± 0.01	0.729 ± 0.017	0.668 ± 0.013
	EGBLUP	0.724 ± 0.016	0.641 ± 0.01	0.721 ± 0.019	0.664 ± 0.014
	BayesA	0.730 ± 0.015	0.658 ± 0.009	0.720 ± 0.016	0.667 ± 0.014
	BayesB	0.746 ± 0.015	0.667 ± 0.009	0.723 ± 0.019	0.670 ± 0.013
	BayesC	0.737 ± 0.015	0.662 ± 0.01	0.726 ± 0.018	0.668 ± 0.014
	deepGBLUP	0.752 ± 0.016	0.673 ± 0.009	0.746 ± 0.017	0.672 ± 0.012
5000	GBLUP	0.682 ± 0.018	0.581 ± 0.009	0.679 ± 0.018	0.609 ± 0.012
	DGBLUP	0.684 ± 0.018	0.576 ± 0.009	0.683 ± 0.019	0.610 ± 0.012
	EGBLUP	0.678 ± 0.017	0.578 ± 0.01	0.676 ± 0.019	0.606 ± 0.013
	BayesA	0.678 ± 0.018	0.581 ± 0.008	0.664 ± 0.017	0.602 ± 0.012
	BayesB	0.697 ± 0.017	0.593 ± 0.009	0.677 ± 0.019	0.606 ± 0.012
	BayesC	0.684 ± 0.018	0.586 ± 0.009	0.673 ± 0.019	0.607 ± 0.012
	deepGBLUP	0.712 ± 0.018	0.607 ± 0.009	0.702 ± 0.018	0.619 ± 0.011
2500	GBLUP	0.631 ± 0.016	0.515 ± 0.011	0.627 ± 0.025	0.539 ± 0.01
	DGBLUP	0.634 ± 0.016	0.514 ± 0.012	0.628 ± 0.024	0.539 ± 0.01
	EGBLUP	0.629 ± 0.016	0.514 ± 0.012	0.625 ± 0.025	0.538 ± 0.01
	BayesA	0.612 ± 0.016	0.500 ± 0.012	0.600 ± 0.022	0.525 ± 0.01
	BayesB	0.635 ± 0.015	0.515 ± 0.012	0.615 ± 0.025	0.531 ± 0.009
	BayesC	0.622 ± 0.016	0.508 ± 0.011	0.615 ± 0.025	0.534 ± 0.009
	deepGBLUP	0.660 ± 0.016	0.544 ± 0.013	0.650 ± 0.023	0.552 ± 0.01
1000	GBLUP	0.532 ± 0.017	0.384 ± 0.02	0.528 ± 0.018	0.424 ± 0.014
	DGBLUP	0.532 ± 0.017	0.381 ± 0.021	0.527 ± 0.018	0.424 ± 0.014
	EGBLUP	0.532 ± 0.017	0.384 ± 0.02	0.527 ± 0.018	0.423 ± 0.014
	BayesA	0.487 ± 0.018	0.361 ± 0.022	0.479 ± 0.018	0.404 ± 0.016
	BayesB	0.502 ± 0.015	0.365 ± 0.019	0.496 ± 0.019	0.405 ± 0.014
	BayesC	0.505 ± 0.015	0.365 ± 0.02	0.510 ± 0.018	0.402 ± 0.016
	deepGBLUP	0.557 ± 0.018	0.432 ± 0.018	0.564 ± 0.019	0.438 ± 0.013

Each value in the cells are means and standard errors of the predictive abilities for 10-fold tests. We highlight the best results in italic