VALIDATIONF90

This program processes solutions obtained with blupf90+ or gibbsf90+ to obtain validation statistics presented in Legarra et al. 2008 and Legarra et al. 2018. Confidence intervals for validation statistics are calculated as in Bermann et al. 2024.

The program works by obtaining solutions with a training (partial) dataset and the complete (whole) dataset. Then, validation statistics are obtained from those solutions and a list of validation individuals.

The steps to do validation with the blupf90 software suite are

• Run renumf90.
• Identify the validation individuals and save a list with their renumbered identifications.
• Create the partial dataset.
• Run blupf90+ or gibbsf90+ to obtain solutions with the whole data. Rename the solutions file to solutions_whole.
• Run blupf90+ or gibbsf90+ to obtain solutions with the partial data. Rename the solutions file to solutions_partial.
• Run validationf90.

OPTION validation eff ids

eff must indicate the effect to validate and ids is the name of a file containing a list of the renumbered identifications of the validation set

OPTION predictive_ability

calculates predictive ability (correlation between predictions and adjusted phenotypes) based on the file yhat_residual obtained from predictf90

OPTION prefix filename

the program reads filename_whole and filename_partial instead of solutions_whole and solutions_partial, respectively

OPTION focal_var mode x

Genetic variance in the validation set. It can be calculated from the inbreeding of the validation set or specified by the user. mode can be inb (default) or usr. When mode is usr, x are the genetic variances in the validation set for each trait

OPTION se type

Calculates standard errors (sd) and confidence intervals (ci) for the validation statistics. type can be:

• exact: frequentist sd and ci are calculated. The prediction error variances matrices for each trait are required. They can be obtained by running blupf90+ with OPTION store_pev_pec eff full, where eff is the effect to validate.
• approx: the frequentist sd and ci are approximated based on reliabilities. Reliabilities of ebv are required. They can be obtained with OPTION store_acc eff from blupf90+ or with accf90/accf90GS/accf90GS2.
• boot: sd and ci are obtained by bootstrapping.