Enhancing the prediction accuracy of groundnut yield by integrating significant markers and modeling genotype × environment interaction.

Multi-environment trials are routinely conducted in plant breeding to capture the genotype-by-environment interaction (G × E) effects. Significant G × E could alter the response pattern of genotypes (the change in rankings of genotypes), subsequently complicating the selection process. Four genomic prediction (GP) models were assessed in three groundnut yield-related traits: pod yield (PY), seed weight (SW), and 100 seed weight (SW100), across four environments. The models, M1 (environment + line), M2 (environment + line + genomic), M3 (environment + line + genomic + genomic × environment interaction), and M4 (environment + line + genomic + genomic × environment interaction + significant markers), were tested using four cross-validation (CV) schemes (CV2, CV1, CV0, and CV00), each simulating different practical breeding scenarios. The results revealed that models incorporating marker data (M2, M3, and M4) consistently improved predictive ability in comparison to the phenotypic model (M1). Incorporating G × E (M3 and M4) further improved predictive ability and reduced residual and environmental variances. The inclusion of significant markers and G × E was more advantageous in CV1 and CV00 scenarios, demonstrating that this strategy is especially useful when phenotypic data for the target genotypes is limited or unavailable. Across the CV schemes, predictive ability was higher in CV2, suggesting that including additional information on the performance of genotypes in known environments can increase the accuracy of selecting superior genotypes in breeding programs. Integrating significant markers and modeling G × E in GP models could be an effective approach in groundnut breeding programs to accelerate genetic gains.