Harnessing epistatic architecture to improve genomic selection accuracy for soybean pod number

Soybean (Glycine max [L.] Merr.) is a globally critical oilseed and protein crop, and pod number per plant (PN) is a core determinant of soybean grain yield. Conventional genomic selection (GS) for PN almost exclusively focuses on additive genetic effects, leading to low prediction accuracy, while traditional genome-wide epistasis analysis suffers from high false positive rates and insufficient biological validation. In this study, based on the Nongdadou2 reference genome and resequencing data of 477 soybean accessions, we performed pathway-level genome-wide epistasis analysis for PN via the BridGE pipeline, identified non-redundant favorable epistatic pairs, constructed epistatic interaction networks, and validated the biological authenticity of these pairs through AlphaFold2-based protein three-dimensional structure modeling and protein-protein interaction enrichment test, along with genotypic combination effect analysis. Integrating these validated favorable epistatic effects into six mainstream GS models achieved an average absolute improvement of 0.42 in PN prediction accuracy across all models. This study provides an effective strategy and practical technical framework for optimizing GS performance for soybean PN.

The online version contains supplementary material available at https://doi.org/10.1007/s11032-026-01711-3.