. 2019 Mar 19;132(7):1943–1952. doi: 10.1007/s00122-019-03327-y

Table 3.

Prediction accuracies using the largest training sets by cross-validation approach

	Correlation metric	Training set size	Correlation^a	Bi-/tri-parental^d
Leave-one-cross-out	By cross	2787	0.127 _0.222	0.12 _0,.20/0.20 _0.11
Tenfold, crosses	By cross	2563	0.125 _0.193	0.11 _0.20/0.20 _0.08
Tenfold, random^b	By cross	2567	0.142 _0.195	0.12 _0.17/0.24 _0.09
Tenfold, crosses	Across all^c	2567	0.289 _0.259	N/A
Tenfold, random^b	Across all^c	2567	0.543 _0.009	N/A

^aAverage across all replicates. Small font displays inter-quantile range for correlations

^bTenfold cross-validation where validation and training sets were grouped by lines instead of crosses

^cCorrelations were calculated across multiple crosses in validation set

^dAverage correlation (^a), but across bi-parental or tri-parental crosses