Maximizing efficiency in sunflower breeding through historical data optimization

Table 1 Predictive ability of GBLUP and GBM models across all traits using all available data for the selected years to calibrate them

Trait	Model	Test Set	Oldest year included in the training set
Trait	Model	Test Set	6	5	4	3	2	1	2*	1*
YLD	GBM	6	NA	0.344	0.370	0.352	0.356	0.356	0.384	0.383
	GBM	7	0.328	0.358	0.374	0.372	0.384	0.383	0.381	0.377
	GBLUP	6	NA	0.368	0.365	0.337	0.344	0.357	0.370	0.380
	GBLUP	7	0.338	0.369	0.374	0.368	0.377	0.373	0.379	0.374
GM	GBM	6	NA	0.418	0.446	0.460	0.463	0.475	0.445	0.453
	GBM	7	0.432	0.467	0.481	0.473	0.475	0.486	0.487	0.502
	GBLUP	6	NA	0.390	0.398	0.420	0.425	0.427	0.401	0.407
	GBLUP	7	0.400	0.436	0.481	0.489	*0.480*	0.490	0.475	0.488
OIL	GBM	6	NA	0.467	0.531	0.548	0.557	0.556	0.549	0.550
	GBM	7	0.419	0.424	0.464	0.467	0.470	0.478	0.467	0.471
	GBLUP	6	NA	0.512	0.542	0.539	0.546	0.551	0.555	0.560
	GBLUP	7	0.420	0.461	0.488	0.489	0.493	0.488	0.495	0.488

The analysis considered various training set-test set combinations, with two test sets representing data for years 6 and 7. The training sets were constructed by including data from the year preceding the test set, two years prior to the test set, and so on until all available data older than the test set was included. We have highlighted in italic all cases where adding an additional year resulted in reduced predictive ability and in bold the year combinations selected by multi-objective optimization. The training sets with an asterisk next to them (Last two columns) indicate that year 3 has been excluded from them. Although these latter two scenarios were not among those initially planned, in light of the results from the multi-objective optimization, we decided to include them in our analysis

ISSN: 1746-4811