Comparative evaluation of GBLUP and PLSR models for genomic prediction of tiller number in rice

Abstract

Genomic prediction (GP) is an advanced approach in modern crop improvement programs that utilizes genome-wide marker data to estimate the breeding values of genotypes. This study compares the predictive performance of two GP models viz., Genomic Best Linear Unbiased Prediction (GBLUP) and Partial Least Squares Regression (PLSR) for estimating the Genomic Estimated Breeding Values (GEB’s) of tiller number (TN) in the genetically diverse Bengal and Assam Aus Panel (BAAP) of rice. The analysis used phenotypic data from over 200 genotypes and approximately two million SNP markers collected under low-nitrogen (N0) conditions during the Kharif 2021 season. Although both models exhibited moderate prediction accuracies, GBLUP outperformed PLSR, achieving a higher coefficient of determination (R² = 0.30) compared to PLSR (R² = 0.13). These results highlight GBLUP's relative advantage for predicting complex traits like tiller number, particularly in resource-constrained breeding programs.