Association mapping analysis of oil palm interspecific hybrid populations and predicting phenotypic values via machine learning algorithms

The genotyping-by-sequencing (GBS) approach was applied to genotype selected interspecific hybrid (F1) and backcross two (BC2) families of Elaeis oleifera and Elaeis guineensis. Genome-wide linkage disequilibrium (LD) was estimated at 150-kb pairwise distance for r2 values of .17 and .42 for the F1 and BC2, respectively. Single marker-trait association analysis revealed that 47 markers were associated with five fatty acid composition (FAC) traits (C16:0, C18:0, C18:1, C18:2 and iodine value [IV]) in F1, and that 12 significant markers were linked to oleic acid (C18:1) and vegetative traits (petiole width and mean leaf width) in BC2. Within the QTL region associated with FAC traits, we identified key candidate genes influencing fatty acid synthesis. We implemented two machine learning algorithms, namely random forest and gradient boosting, to evaluate the ability of significant markers in predicting phenotype values. We also demonstrated the contribution of different marker combinations on trait values via prediction trees. This is the first attempt to evaluate the predictive ability of a combination of markers associated with traits identified from association mapping analysis in oil palm populations.