Ensemble Methods for Selecting Single Nucleotide Polymorphisms Associated to Rice Phenotypes

Rice (Oryza sativa) is one of the largest collections of genetic resources among plant species of economic interest. Several genetic variability studies have been developed to increase this cultivar’s productivity. In this context, Single Nucleotide Polymorphisms (SNPs), single base variations in DNA sequences, have been widely studied, as they act as molecular markers linked to productivity and resistance in rice cultivation. Due to the ineffectiveness of conventional methods in selecting SNPs, methods based on Machine Learning have been used. For this purpose, the selection of SNPs is modeled as a Feature Selection problem. Although feature selection is widespread in the literature, there are still gaps regarding its use in the context of rice genetic improvement. To advance interesting points regarding this discussion, we propose two ensemble methods for selecting important SNPs related to different phenotypes in rice, combining feature selection algorithms to generate a robust result. Our first proposal directly selects the most important SNPs using the phenotype numeric values. The second proposal discretizes the phenotype values, selecting the most important SNPs through classification algorithms. Experiments using real-world rice datasets showed that the proposed ensembles were better or very competitive compared to other methods from the literature regarding SNPs selected and prediction performance.