Assessing the Potential of Near-Infrared Spectroscopy for Predicting Yield Component Traits in Sweetpotato

Near-infrared spectroscopy (NIRS) is routinely employed in sweetpotato breeding programs at the International Potato Center (CIP) as a way to assess storage root quality traits, including carotenoid, sugar, and mineral contents. As such, NIRS-based phenomic prediction may provide an affordable alternative to genomic prediction for accelerating genetic gains in the crop. In this study, we investigated the potential of NIRS in predicting yield-related traits and compared its accuracy with genomic prediction in a multiparental sweetpotato population from CIP-Peru. The 720 clones comprise six biparental populations from a partial diallel cross of eight parents, representing two putative heterotic groups. Predictive ability was assessed using four cross-validation schemes (CV2, CV1, CV0, and CV00). Pedigree-BLUP outperformed Genomic-BLUP and all NIRS-based models for predicting number of commercial roots (NOCR) under CV2, in which the tested genotypes were evaluated in tested environments. However, for CV1, CV0, and CV00, NIRS-BLUP showed higher predictive ability for NOCR. NIRS-BLUP consistently outperformed G-BLUP and Ped-BLUP for root yield (RYTHA) and harvest index (HI) across the four cross-validation schemes. The moderate to high predictive ability observed indicates the potential of NIRS-based phenomic prediction as a cost-effective alternative to predicting unobserved phenotype in sweetpotato.