Identification of genetic markers for increasing agricultural productivity: An empirical study

Abstract

Genomic selection (GS) has been used globally for increasing agricultural production and productivity. It has been used for complex quantitative traits by selecting breeding material after predicting Genomic Estimated Breeding Values (GEBVs) of target species. The accuracy of GS for estimation of GEBVs depends on various factors including sampling population, genetic architecture of target species, statistical models, etc. The feature (marker) selection is one of the important steps in development of GS models. There are large numbers of models proposed in the literature for GS. However, applicability of these models is based on many factors including extent of additive and epistatic effects of breeding population. Therefore, there is strong need to evaluate the performance of these models and techniques of feature selection under different situations. In this study, performance of linear/additive effect models, viz. linear least squared regression, BLUP, LASSO, ridge regression, SpAM as well as non-linear/epistatic effect models, viz. mRMR, HSIC LASSO have been evaluated through a simulation study in R platform. In general, performance of SpAM was found to be superior for GS than all other models considered in this study in case of presence of additive effect and absence of epistatic effect. However, in case of low heritability and high epistatic effect the HSIC LASSO outperformed all models. This study will assist researcher in selection of appropriate feature selection technique for a given situation.