GWAS and Regularised Regression Identify SNPs Associated with Candidate Genes for Stage-Specific Salinity Tolerance in Rice

Soil salinity remains a major constraint to rice productivity, particularly during early developmental stages when plants are highly sensitive to osmotic and ionic stress. In this study, we evaluated 201 genetically diverse rice genotypes from the 3K Rice Diversity Panel to investigate stage-specific mechanisms of salinity tolerance and develop machine learning-based predictive models for rapid phenotypic screening. Morphological and physiological traits were measured under control and saline conditions at germination and early seedling stages to derive Stress Tolerance Indices (STIs). The average membership function value (AMFV), calculated from multi-trait STI profiles, effectively captured variation in salinity responses and enabled classification of genotypes into five tolerance categories. Genome-wide association analysis using high-density SNP markers identified 36 significant marker–trait associations, including potentially novel SNPs on chromosomes 1 and 12. Several loci co-localized with candidate genes (LTR1, LGF1, OsCPS4, OsNCX7, and OsNHX4), while functional SNPs within genes (OsDRP2C, RLCK168, and OsMed37_2) and non-synonymous variants (qSVII11.1 and qSNaK3.1) further supported their candidacy in salinity tolerance. Mining favourable SNPs of causal genes identified superior multilocus combinations consistent with STI-based phenotypic patterns, with genotype 91-382 emerging as the strongest performer, exhibiting enhanced Na+ exclusion, K+ retention, and biomass resilience across developmental stages. To address multicollinearity among STI traits, we applied cross-validated LASSO (germination) and Elastic Net (early seedling) models, achieving high predictive accuracy and revealing a developmental shift from biomass-driven tolerance at germination to ion-regulatory processes at the seedling stage. Independent validation showed strong agreement between predicted and observed AMFVs. By integrating physiological indices, GWAS-derived SNP signals, and regularized machine learning approaches, this study provides a robust framework for identifying elite donors and accelerating breeding for salt-tolerant rice.