The Use of a Simple Genetic Algorithm as a Tree Breeding Selection Technique

Abstract

SYNOPSIS A genetic ranking procedure for breeding which is not dependant on statistical selection methods such as Best Linear Unbiased Prediction (BLUP) and Best Linear Prediction (BLP) is described and results of tests on simulation breeding data are presented. A simple genetic algorithm (SGA) is adapted for the purposes of predicting genetic breeding values of individuals in a population. Coefficients are used to pool the sources of information (such as individual and family values). The coefficients are generated by the SGA, which selects the coefficients by means of a fitness test of the genetic gains (of the potential prediction coefficients) realised in a simulation population. This type of SGA solution realised on average significantly greater gains than did BLP in simulation tests. The genetic gains the SGA realised in the simulation studies did not surpass the predicted gains of BLP, but BLP did not realise the predicted gains in 80 percent of the simulation cases. Instability described in classical regression literature may also be present in Best Linear Prediction (BLP) and Best Linear Unbiased Prediction (BLUB).