Prediction accuracy of genomic estimated breeding values for fruit traits in cultivated tomato (Solanum lycopersicum L.)

Table 2 Prediction accuracy of cross-validation methods for five fruit traits in three tomato training populations

Training population	Method^a	Training set size	Prediction accuracy^b
Training population	Method^a	Training set size	FW^c	FWt	FH	PT	Brix
TGC1 (n = 162)	LOOCV	161	0.822	0.867	0.821	0.856	0.670
	k-fold	145 (k = 10)	0.823	0.859	0.813	0.847	0.636
	k-fold	129 (k = 5)	0.807	0.853	0.806	0.851	0.652
TGC2 (n = 191)	LOOCV	190	0.748	0.766	0.687	0.618	0.776
	k-fold	171 (k = 10)	0.747	0.765	0.698	0.618	0.747
	k-fold	152 (k = 5)	0.741	0.762	0.687	0.614	0.761
Combined (n = 353)	LOOCV	352	0.758	0.802	0.719	0.765	0.736
	k-fold	317 (k = 10)	0.754	0.798	0.715	0.752	0.723
	k-fold	282 (k = 5)	0.741	0.790	0.703	0.748	0.727

^aTwo cross-validation methods, leave-one-out cross-validation (LOOCV) and k-fold (k = 10 and 5) were evaluated and each k were iterated in 100 different dividing patterns
^bPrediction accuracy was estimated using the Pearson correlation coefficients between genomic estimated breeding values (GEBVs) and observed phenotypes. The GEBVs were calculated using the confident 31,142 SNPs in the RR-BLUP model
^cFW (fruit weight), FWt (fruit width), FH (fruit height), and PT (pericarp thickness)

ISSN: 1471-2229