I have been reading several documents in which methodologies are compared in terms of accuracy: the correlation between true and predicted genetic values. Higher accuracy, we trust more the results; we are more certain of making a good selection.
A funny aspect of how we think is the assumption that if the selection has low accuracy then the performance of the genotypes will be lower than predicted. However, it is also likely that performance will be higher than predicted; we are just not very certain about it.
We can think of our breeding programmes as having two objectives:
- Shifting the population average for one or, more commonly, more traits and
- Identifying a small subset of genotypes that will be commercially deployed at large scale.
Objective 1 is quite resistant to low accuracies; if we select a ‘dud’ (we got the negative tail of low accuracy), we can get rid of it for further breeding. We could also get lucky (the positive tail of low accuracy) and get a very superior genotype.
Objective 2 needs more pause and higher accuracy. We do not want to bet the farm on the chance of getting a positive tail, but to inspire confidence on the material that is delivered by the breeding programme. Therefore, low and high accuracies can coexist within a programme if we remember that each of them has their particular use.
