Genomic selection for key traits affecting biological yield of maize (Zea mays L.) under optimal phosphorus and phosphorus deficient conditions

Phosphorus deficiency is one of the most severe abiotic constraints limiting maize ( Zea mays L.) productivity, particularly in semi-arid regions. Genomic selection (GS) has emerged as a promising strategy for improving complex agronomic traits under stress conditions. The present study aimed to assess the efficiency of GS models in predicting key agronomic traits of maize under both optimal phosphorus supply and phosphorus-deficient environments. A panel of 93 maize genotypes was phenotyped for 34 agronomic traits, and using artificial neural network analysis, eight key traits under optimal phosphorus conditions and five key traits under phosphorus-deficient conditions were identified as the most influential contributors to biological yield and subsequently used for genomic selection, and genotyping was performed using high-density SNP markers. Genomic estimated breeding values (GEBVs) were predicted using seven genomic prediction models, including Bayes A, Bayes B, Bayes C, Bayesian Lasso (BL), Bayesian Ridge Regression (BRR), Genomic Best Linear Unbiased Prediction (GBLUP), and Ridge Regression Best Linear Unbiased Prediction (RRBLUP). Among the evaluated models, Bayesian approaches, particularly Bayes A, generally exhibited higher correlations between phenotypic values and genomic estimated breeding values (GEBVs), whereas BL showed comparatively lower performance. Phenotypic–genotypic correlations ranged from 0.85 to 0.96 under optimal conditions and from 0.61 to 0.95 under phosphorus-deficient stress. Heritability estimates were moderate to high across traits. Principal component analysis explained over 99% of the total variance, indicating substantial genetic diversity and strong model convergence. Cluster analysis classified genotypes into three distinct groups, with Ma010, Ma001, Ma011, and Ma013 consistently exhibiting favorable GEBVs across both phosphorus regimes. The results indicate that Bayesian models generally provided the most favorable overall performance across the evaluated traits, highlighting their potential utility in genomic selection programs targeting phosphorus-deficiency tolerance. The integration of GS with advanced statistical and machine learning approaches provides a robust framework for accelerating the identification of superior, stress-tolerant maize genotypes and supports the development of high-yielding cultivars adapted to nutrient-deficient environments.