Appliaction of classification algorithms for Genomic Prediction methods in an intraspecific cross of Eucalyptus grandis

MARTÍN GARCÍA; LEONARDO ORNELLA; EDUARDO CAPPA; PAMELA VILLALBA; CINTIA ACUÑA; CAROLINA MARTINEZ; MAURO SURENCISKI; JAVIER OBERSCHELP; LEONEL HARRAND; JUAN LÓPEZ; ELIZABETH TAPIA; SUSANA MARCUCCI POLTRI

Rosario, Argentina

Congreso; 4° Congreso Argentino de Bioinformática y Biología Computacional (4CAB2C); 2013

In a previous report we present the results of the performance of Bayes LASSO, Ridge Regression, Reproducing Kernel Hilbert Spaces (RKHS) and Random Forest (RFR) applied on a dataset of a 131 full sibs of Eucalyptus grandis [1]. Eventhought results were relatively good ( ranging from 0.205 to 0.767), by graphical analysis we observed an important drop in the accuracy of prediction of individuals belonging the extremes of the distribution, i.e., where the breeder selects the individuals for the next generation.Classification algorithms are a very successful branch of supervised machine learning; they are fully applied in several areas of research (from text mining to bioinformatics) but we found very few studies on its application in genomic selection.Here we compare the performance of regression algorithms with classification methods: Random Forest Classification (RFC) and Support Vector Classification with linear kernel (SVC-lin): instead of building a regression curve that fits all the training data, (binary) classification algorithms construct a decision frontier that is optimized for separating usually two classes, i.e., the best and worst candidates; thus we expected they might be a substitute approach for regression algorithms.