MGIDI selection and machine learning reveal harvest index driving traits in sodium azide–induced rice mutants with SSR-based genetic diversity

Abstract Sodium azide mutagenesis offers a powerful approach to generate genetic diversity for rice improvement, yet comprehensive characterization of mutant populations using integrated modern breeding tools remains limited. M₃ mutants of BRRI dhan28 induced with sodium azide, were evaluated for 17 agronomic traits and genetic diversity was characterized using 30 SSR markers. The MGIDI was used to characterize elite genotypes and machine learning approaches were used to dissect trait architecture underlying harvest index. The phenotypic variation captured by principal component analysis was 52.12%, and yield was the trait with the highest genotypic variance (278.22) and genotypic coefficient of variation (29.07%). MGIDI analysis detected 10 elite mutants that significantly outperformed within the same environment in combined yield and harvest index. The main predictors of harvest index variability were examined using a Random Forest analysis, and this showed that grain and straw yield were the main predictors of harvest index variability. The SSR markers showed high level of genetic diversity (PIC = 0.264), population structure analysis revealed two subgroups (Fst = 0.0437) and the pairwise genetic distance ranged from 0.000 to 0.733. Procrustean alignment showed a high correlation between molecular and phenotypic variation. An integrated approach of MGIDI selection and prediction of diversity using machine learning underpinned the identification of elite mutants that can be quickly forwarded to breeding programs. This study provides valuable genetic resources and demonstrates that sodium azide mutagenesis combined with modern analytical tools accelerates genetic gains in rice improvement. Competing Interest Statement The authors have declared no competing interest.