Efficacy of gaussian mixture models for genotype selection in coffee bean

Abstract

Coffee is one of the most important traded commodities in the world. It is well known that coffee bean yield is subjected to strong variation through the years in a phenomenon called biennial growth. This behavior has imposed great challenges on statistical analysis of coffee bean yield data. In these species genotypes show a differential biennial behavior due to its physiological response to environmental condition which suggests a mixture of subpopulations. Previous studies have tried to solve the problem, however they assume the presence of only one stochastic process generating the phenotypes. In the first paper it is proposed a finite mixture model to deal with the biennial pattern as hidden variable. Individual (per harvest) and repeated measures analyses were performed using conventional mixed models and Gaussian mixture mixed models. The results showed a great increase on parameter efficiency estimation and lead to greater genetic gain suggesting that for analysis of C. arabica progenies exhibiting different biennial patterns, mixture mixed models are superior to traditional mixed models and to models that structure biennial effects using covariance matrices. On the second paper the gaussian mixed mixture model is extended for genomic prediction (GMGBLUP) and compared with a traditional genomic prediction model (GBLUP). The aim was to verify the prediction accuracy when the markers effects are corrected for bias of the biennial growth. For the real data set the GBLUP performed better in all harvests, however the simulated data results showed that the GMGBLUP is superior when the subpopulations means are contrasting and the mixture parameter is close to 0.5. The results suggest that GMGBLUP should be considered as an alternative for genomic prediction in coffea genus, especially for species with strong biennial growth behavior.