Mining genes, genomic selection and simulation breeding on hundred-seed weight using a four-way RIL population

Hundred-seed weight (HSW) is a critical determinant of soybean yield potential, yet its genetic dissection is often limited by the restricted allelic diversity of traditional biparental populations. In this study, we developed a four-way recombinant inbred line (FW-RIL) population comprising 144 lines derived from four founders with distinct HSW phenotypes to enhance mapping resolution and allelic richness. By integrating a high-density genetic linkage map with five multi-locus genome-wide association study (GWAS) models, we systemically analyzed the genetic basis of HSW across multiple environments. A total of 16 quantitative trait loci (QTLs) and 40 quantitative trait nucleotides (QTNs) were identified, among which three QTLs and two QTNs were consistently detected across environments and analytical approaches. Co-localization analysis, combined with linkage disequilibrium (LD) mapping, haplotype effect evaluation, and gene expression profiling, prioritized Glyma.02G047200 as the key candidate gene within the stable major-effect locus qHSW-2-1. This gene encodes an oligopeptide transporter, and haplotype analyses demonstrated that elite alleles significantly increase seed weight across diverse genetic backgrounds. Furthermore, comparative evaluation of multiple genomic selection (GS) models revealed that the Light Gradient Boosting Machine (LightGBM) achieved the highest predictive accuracy for HSW. Using this model, breeding simulations identified several high-yielding hybrid combinations pyramiding multiple superior allelic variants. Overall, this study elucidates the genetic determinants of soybean HSW and highlights the effectiveness of integrating multi-parent populations, genomic selection, and molecular design breeding to accelerate yield improvement.