Joint genetic control of isoflavones and soyasaponins revealed by mGWAS, genomic prediction, and SHAP-guided allele stacking

Abstract

Isoflavones and soyasaponins are two classes of health-promoting specialized metabolites in soybean, and improving them simultaneously is a key breeding goal. Emerging evidence indicates that these two metabolite classes can act synergistically in vivo and in vitro, making their simultaneous enhancement an increasingly important breeding objective. However, despite extensive studies on each pathway independently, the genetic basis underlying joint variation of isoflavones and soyasaponins remains poorly understood. Here, we profiled 17 metabolites (12 isoflavones and 5 soyasaponins) across 376 accessions of the Korean soybean core collection using UPLC. We characterized metabolite distributions, correlations, and presence–absence patterns, and performed multi-metabolite Genome-Wide Association Study (GWAS), identifying 70 high-confidence loci. These included previously reported major loci as well as eight novel loci for isoflavones and thirteen for soyasaponins. Five genomic regions showed shared linkage disequilibrium (LD) structure between the two pathways, and we identified candidate genes for high-confidence loci. We next compared FT-IR–based phenomic prediction with GWAS-informed genomic prediction, finding that genomic prediction consistently outperformed phenomic prediction and achieved moderate to high accuracy, indicating strong genetic determinism. Finally, we applied an XGBoost– SHapley Additive exPlanations (SHAP) framework to estimate the extent to which favorable alleles could be combined in silico. Single-trait allele stacking pointed to CMJ_115, CMJ_068, and CMJ_236 as the best-performing accessions for Acetyl-daidzin, Malonyl-daidzin, and Soyasaponin-ab, respectively. Multi-trait optimization produced a virtual genotype most similar to CMJ_317, suggesting this accession as a practical parent for jointly improving both metabolite classes. Overall, our findings provide a population-scale map of diversity, genetic factors, and achievable breeding gains for functional soybean improvement.