Genomic prediction in dolichos bean (Lablab purpureus L. Sweet) breeding: choice of the type of training population and selection generation

Dolichos bean is a nutritionally rich yet underutilized orphan legume crop. Conventional phenotype-based genetic improvement has been slow; therefore, integrating genomic selection (GS) into breeding programs is essential to accelerate genetic gain. Our objective in this study was to optimize the factors influencing GS success in dolichos bean by identifying: (i) the most effective GS model, (ii) an appropriate training population (TP), and (iii) the ideal selection generation for fresh pod yield. Four populations with distinct genetic structures were evaluated for fresh pod yield and genotyped using 413 SSR markers distributed across 11 linkage groups. The TPs were evaluated across five environments (locations and/or years), whereas the breeding populations (BPs) were evaluated over two seasons. Four GS models were trained on two differentially structured TPs and validated on two breeding populations. The optimal generation for genomic prediction was assessed by comparing an early-generation segregating population (F 2:3 ) with an advanced-generation population (F 5:6 ). Our results indicate that the RKHS model showed the highest predictive ability (i.e. correlation between observed and predicted fresh pod yield) compared to the other three models. We found that a closely structured Recombinant Inbred Line (RIL) population is the most effective for training genomic prediction models. Furthermore, predictive abilities were comparable between the F 2:3 and F 5:6 generations. We conclude that predicting fresh pod yield in early segregating generations using an RKHS model trained on a RIL population is highly effective strategy for implementing GS in dolichos bean improvement programs.