Multivariate Genomic Best Linear Unbiased Prediction (GBLUP) Can Improve Genomic Prediction in Mungbean

Genomic prediction (GP) remains underexplored in mungbean despite its potential to accelerate genetic gain in breeding programs. This study evaluated the implementation of GP in mungbean using multivariate genomic best linear unbiased prediction (GBLUP), including multi-trait, multi-environment and multi-trait multi-environment models to maximize prediction accuracy. Utilizing yield (YLD), days-to-50%-flowering (DTF), and plant height (PHT) data collected from 313 diverse mungbean accessions (including accessions released and widely grown in Australia) tested across four environments and genotyped for 3,310 SNPs, we compared multivariate GBLUP models including single-trait−multi-environment, multi-trait−single-environment, and multi-trait−multi-environment against univariate single-trait−single-environment models. For multi-trait models, YLD was the primary trait, while DTF and PHT were the secondary traits. Secondary traits are correlated traits which are easy and relatively inexpensive to measure early in crop development. Cross-validation schemes implemented assessed whether secondary traits were available at the time of prediction of YLD for the validation sets (no secondary traits, NOSEC vs secondary traits available, SEC) for multi-trait models, cross-environment data availability (no cross-environment data, Scenario 1 vs cross-environment data available, Scenario 2) for multi-environment models and their factorial combinations (NOSEC/SEC × Scenario1/2) for multi-trait multi-environment models could improve prediction accuracy. Single-trait single-environment GBLUP models provided baseline prediction accuracies of 0.33 ± 0.02 to 0.61 ± 0.01 for YLD, 0.39 ± 0.02 to 0.49 ± 0.02 for DTF, and 0.38 ± 0.01 to 0.59 ± 0.01 for PHT. Multivariate approaches substantially improved prediction accuracy, with gains up to + 52.7% for YLD, + 100.22% for DTF and + 83.43% for PHT, over baseline models. The magnitude of improvement depended on trait heritability, genetic correlation between traits and between environments, and availability of information on the validation set. These findings demonstrate that genomic prediction is feasible in mungbean and can be substantially enhanced through multivariate approaches, providing a foundation for implementing genomic selection in mungbean breeding programs.